Search results for: spoken word recognition

Authors: Milind Patil

Abstract:

Present paper is a discussion on Hindi-Marathi Morphological Analysis and generating rules for Machine Translation on the basis of Machine Readable Dictionary (MRD). This used Transformative Generative Grammar (TGG) rules to design the MRD. As per TGG rules, the suffix of a particular root word is based on its Tense, Aspect, Modality and Voice. That's why the suffix is very important for the word meanings (or root meanings). The Hindi and Marathi Language both have relation with Indo-Aryan language family. Both have been derived from Sanskrit language and their script is 'Devnagari'. But there are lots of differences in terms of semantics and grammatical level too. In Marathi, there are three genders, but in Hindi only two (Masculine and Feminine), the Natural gender is absent in Hindi. Likewise other grammatical categories also differ in their level of use. For MRD the suffixes (or Morpheme) are of particular root word for GNP (Gender, Number and Person) are based on its natural phenomena. A particular Suffix and Morphine change as per the need of person, number and gender. The design of MRD also based on this format. In first, Person, Number, Gender and Tense are key points than root words and suffix of particular Person, Number Gender (PNG). After that the inferences are drawn on the basis of rules that is (V.stem) (Pre.T/Past.T) (x) + (Aux-Pre.T) (x) → (V.Stem.) + (SP.TM) (X).

Keywords: MRD, TGG, stem, morph, morpheme, suffix, PNG, TAM&V, root

Procedia PDF Downloads 295

Authors: M. Jazouli, S. Elhoufi, A. Majda, A. Zarghili, R. Aalouane

Abstract:

Autism spectrum disorder is a complex developmental disability. It is defined by a certain set of behaviors. Persons with Autism Spectrum Disorders (ASD) frequently engage in stereotyped and repetitive motor movements. The objective of this article is to propose a method to automatically detect this unusual behavior. Our study provides a clinical tool which facilitates for doctors the diagnosis of ASD. We focus on automatic identification of five repetitive gestures among autistic children in real time: body rocking, hand flapping, fingers flapping, hand on the face and hands behind back. In this paper, we present a gesture recognition system for children with autism, which consists of three modules: model-based movement tracking, feature extraction, and gesture recognition using artificial neural network (ANN). The first one uses the Microsoft Kinect sensor, the second one chooses points of interest from the 3D skeleton to characterize the gestures, and the last one proposes a neural connectionist model to perform the supervised classification of data. The experimental results show that our system can achieve above 93.3% recognition rate.

Keywords: ASD, artificial neural network, kinect, stereotypical motor movements

Procedia PDF Downloads 280

Authors: Genevieve Jorolan-Quintero

Abstract:

The recording as well as the preservation of indigenous literature is an important task as it deals with a significant heritage of pre-colonial culture. The beliefs and traditions of a people are reflected in their oral narratives, such as the folk epic, which must be written down to insure their preservation. The epic poem for instance, known as dawot among the Mandaya, one of the indigenous communities in the southern region of the Philippines, narrates the customs, the ways of life, and the adventures of an ancient people. Nabayra, an expert on Philippine folkloric studies, stresses that still extant after centuries of unknown origin, the dawot was handed down to the magdadawot (bard) by word of mouth, forming the greatest bulk of Mandaya oral tradition. Unhampered by modern means of communication to distract her/him, the magdadawot has a sharp memory of the intricacies of the ancient art of chanting the panayday (verses) of the epic poem. The dawot has several hullubaton (episodes), each of which takes several nights to chant . The language used in these oral traditions is archaic Mandaya, no longer spoken or clearly understood by the present generation. There is urgency to the task of recording and writing down what remain of the epic poem since the singers and storytellers who have retained the memory and the skill of chanting and narrating the dawot and other forms of oral tradition in their original forms are getting fewer. The few who are gifted and skilled to transmit these ancient arts and wisdom are old and dying. Unlike the other Philippine epics (i.e. the Darangen, the Ulahingan, the Hinilawod, etc.), the Mandaya epic is yet to be recognized and given its rightful place among the recorded epics in Philippine Folk Literature. The general aim of this study was to put together and preserve an intangible heritage, the Mandaya hullubaton (episodes of the dawot), in order to preserve and promote appreciation for the oral traditions and cultural legacy of the Mandaya. It was able to record, transcribe, and translate four hullubaton of the folk epic into two languages, Visayan and English to insure understanding of their contents and significance among non-Mandaya audiences. Evident in the contents of the episodes are the cultural practices, ideals, life values, and traditions of the ancient Mandaya. While the conquests and adventures of the Mandaya heroes Lumungtad, Dilam, and Gambong highlight heroic virtues, the role of the Mandaya matriarch in family affairs is likewise stressed. The recording and the translation of the hullubaton and the dawot into commonly spoken languages will not only promote knowledge and understanding about their culture, but will also stimulate in the members of this cultural community a sense of pride for their literature and culture. Knowledge about indigenous cultural system and philosophy derived from their oral literature will serve as a springboard to further comparative researches dealing with indigenous mores and belief systems among the different tribes in the Philippines, in Asia, in Africa, and other countries in the world.

Keywords: Dawot, epic poem, Mandaya, Philippine folk literature

Procedia PDF Downloads 384

Authors: Tokihiko Akita, Seiichi Mita

Abstract:

A millimeter-wave radar is the most robust against adverse environments, making it an essential environment recognition sensor for automated driving. However, the reflection signal is sparse and unstable, so it is difficult to obtain the high recognition accuracy. Deep learning provides high accuracy even for them in recognition, but requires large scale datasets with ground truth. Specially, it takes a lot of cost to annotate for a millimeter-wave radar. For the solution, utilizing a simulator that can generate an annotated huge dataset is effective. Simulation of the radar is more difficult to match with real world data than camera image, and recognition by deep learning with higher-order features using the simulator causes further deviation. We have challenged to improve the accuracy of traffic participant classification by fusing simulator and real-world data with domain adaptation technique. Experimental results with the domain adaptation network created by us show that classification accuracy can be improved even with a few real-world data.

Keywords: millimeter-wave radar, object classification, deep learning, simulation, domain adaptation

Procedia PDF Downloads 64

Authors: Tatiana N. Astakhova, Alexander E. Saprygin, Tatyana A. Golovko, Alexander N. Savostyanov, Mikhail S. Vlasov, Natalia V. Borisova, Alexandera G. Karpova, Urana N. Kavai-ool, Elena D. Mokur-ool, Nikolay A. Kolchanov, Lubomir I. Aftanas

Abstract:

The aim of the study is to compare behaviorally and EEG reactions in Turkic-speaking inhabitants of Siberia (Tuvinians and Yakuts) and Russians during the recognition of syntax errors in native and foreign languages. 63 healthy aboriginals of the Tyva Republic, 29 inhabitants of the Sakha (Yakutia) Republic, and 55 Russians from Novosibirsk participated in the study. All participants completed a linguistic task, in which they had to find a syntax error in the written sentences. Russian participants completed the task in Russian and in English. Tuvinian and Yakut participants completed the task in Russian, English, and Tuvinian or Yakut, respectively. EEG’s were recorded during the solving of tasks. For Russian participants, EEG's were recorded using 128-channels. The electrodes were placed according to the extended International 10-10 system, and the signals were amplified using ‘Neuroscan (USA)’ amplifiers. For Tuvinians and Yakuts EEG's were recorded using 64-channels and amplifiers Brain Products, Germany. In all groups 0.3-100 Hz analog filtering, sampling rate 1000 Hz were used. Response speed and the accuracy of recognition error were used as parameters of behavioral reactions. Event-related potentials (ERP) responses P300 and P600 were used as indicators of brain activity. The accuracy of solving tasks and response speed in Russians were higher for Russian than for English. The P300 amplitudes in Russians were higher for English; the P600 amplitudes in the left temporal cortex were higher for the Russian language. Both Tuvinians and Yakuts have no difference in accuracy of solving tasks in Russian and in their respective national languages (Tuvinian and Yakut). However, the response speed was faster for tasks in Russian than for tasks in their national language. Tuvinians and Yakuts showed bad accuracy in English, but the response speed was higher for English than for Russian and the national languages. With Tuvinians, there were no differences in the P300 and P600 amplitudes and in cortical topology for Russian and Tuvinian, but there was a difference for English. In Yakuts, the P300 and P600 amplitudes and topology of ERP for Russian were the same as Russians had for Russian. In Yakuts, brain reactions during Yakut and English comprehension had no difference and were reflected foreign language comprehension -while the Russian language comprehension was reflected native language comprehension. We found out that the Tuvinians recognized both Russian and Tuvinian as native languages, and English as a foreign language. The Yakuts recognized both English and Yakut as a foreign language, only Russian as a native language. According to the inquirer, both Tuvinians and Yakuts use the national language as a spoken language, whereas they don’t use it for writing. It can well be a reason that Yakuts perceive the Yakut writing language as a foreign language while writing Russian as their native.

Keywords: EEG, language comprehension, native and foreign languages, Siberian inhabitants

Procedia PDF Downloads 512

Authors: Gamze Uslu, Sebnem Baydere, Alper K. Demir

Abstract:

In this study, data loss tolerance of Support Vector Machines (SVM) based activity recognition model and multi activity classification performance when data are received over a lossy wireless sensor network is examined. Initially, the classification algorithm we use is evaluated in terms of resilience to random data loss with 3D acceleration sensor data for sitting, lying, walking and standing actions. The results show that the proposed classification method can recognize these activities successfully despite high data loss. Secondly, the effect of differentiated quality of service performance on activity recognition success is measured with activity data acquired from a multi hop wireless sensor network, which introduces high data loss. The effect of number of nodes on the reliability and multi activity classification success is demonstrated in simulation environment. To the best of our knowledge, the effect of data loss in a wireless sensor network on activity detection success rate of an SVM based classification algorithm has not been studied before.

Keywords: activity recognition, support vector machines, acceleration sensor, wireless sensor networks, packet loss

Procedia PDF Downloads 447

Authors: Rahma Abed, Sahbi Bahroun, Ezzeddine Zagrouba

Abstract:

Due to the huge amount of data in videos, extracting the relevant frames became a necessity and an essential step prior to performing face recognition. In this context, we propose a method for extracting keyframes from videos based on face quality and deep learning for a face recognition task. This method has two steps. We start by generating face quality scores for each face image based on the use of three face feature extractors, including Gabor, LBP, and HOG. The second step consists in training a Deep Convolutional Neural Network in a supervised manner in order to select the frames that have the best face quality. The obtained results show the effectiveness of the proposed method compared to the methods of the state of the art.

Keywords: keyframe extraction, face quality assessment, face in video recognition, convolution neural network

Procedia PDF Downloads 197

Authors: Cemil Kutluturk

Abstract:

This is a comparative study of the incarnation of God according to Hinduism and Christianity. After dealing with their basic ideas on the concept of the incarnation of God, the main similarities and differences between each other will be examined by quoting references from their sacred texts. In Hinduism, the term avatara is used in order to indicate the concept of the incarnation of God. The word avatara is derived from ava (down) and tri (to cross, to save, attain). Thus avatara means to come down or to descend. Although an avatara is commonly considered as an appearance of any deity on earth, the term refers particularly to descents of Vishnu. According to Hinduism, God becomes an avatara in every age and entering into diverse wombs for the sake of establishing righteousness. On the Christian side, the word incarnation means enfleshment. In Christianity, it is believed that the Logos or Word, the Second Person of Trinity, presumed human reality. Incarnation refers both to the act of God becoming a human being and to the result of his action, namely the permanent union of the divine and human natures in the one Person of the Word. When the doctrines of incarnation and avatara are compared some similarities and differences can be found between each other. The basic similarity is that both doctrines are not bound by the laws of nature as human beings are. They reveal God’s personal love and concern, and emphasize loving devotion. Their entry into the world is generally accompanied by extraordinary signs. In both cases, the descent of God allows for human beings to ascend to God. On the other hand, there are some distinctions between two religious traditions. For instance, according to Hinduism there are many and repeated avataras, while Christ comes only once. Indeed, this is related to the respective cyclic and linear worldviews of the two religions. Another difference is that in Hinduism avataras are real and perfect, while in Christianity Christ is also real, yet imperfect; that is, he has human imperfections, except sin. While Christ has never been thought of as a partial incarnation, in Hinduism there are some partial and full avataras. The other difference is that while the purpose of Christ is primarily ultimate salvation, not every avatara grants ultimate liberation, some of them come only to save a devotee from a specific predicament.

Keywords: Avatara, Christianity, Hinduism, incarnation

Procedia PDF Downloads 227

Authors: T. Damak, O. Kriaa, A. Baccar, M. A. Ben Ayed, N. Masmoudi

Abstract:

In the last few years, Automatic Number Plate Recognition (ANPR) systems have become widely used in the safety, the security, and the commercial aspects. Forethought, several methods and techniques are computing to achieve the better levels in terms of accuracy and real time execution. This paper proposed a computer vision algorithm of Number Plate Localization (NPL) and Characters Segmentation (CS). In addition, it proposed an improved method in Optical Character Recognition (OCR) based on Deep Learning (DL) techniques. In order to identify the number of detected plate after NPL and CS steps, the Convolutional Neural Network (CNN) algorithm is proposed. A DL model is developed using four convolution layers, two layers of Maxpooling, and six layers of fully connected. The model was trained by number image database on the Jetson TX2 NVIDIA target. The accuracy result has achieved 95.84%.

Keywords: ANPR, CS, CNN, deep learning, NPL

Procedia PDF Downloads 280

Authors: Shruti Hajirnis

Abstract:

In May 2014, FASB issued Accounting Standards Update (ASU) 2014-09, Revenue from Contracts with Customers (Topic 606), and the International Accounting Standards Board (IASB) issued International Financial Reporting Standards (IFRS) 15, Revenue from Contracts with Customers that will supersede virtually all revenue recognition requirements in IFRS and US GAAP. FASB and the IASB have basically achieved convergence with these standards, with only some minor differences such as collectability threshold, interim disclosure requirements, early application and effective date, impairment loss reversal and nonpublic entity requirements. This paper discusses the impact of five-step model prescribed in new revenue standard on the entities operating in Pharma industry. It also outlines the considerations for these entities while implementing the new standard.

Keywords: revenue recognition, pharma industry, standard, requirements

Procedia PDF Downloads 420

Authors: Fidel Lòpez Saca, Carlos Avilés-Cruz, Miguel Magos-Rivera, José Antonio Lara-Chávez

Abstract:

Automated object recognition and identification systems are widely used throughout the world, particularly in assembly lines, where they perform quality control and automatic part selection tasks. This article presents the design and implementation of an object recognition system in an assembly line. The proposed shapes-color recognition system is based on deep learning theory in a specially designed convolutional network architecture. The used methodology involve stages such as: image capturing, color filtering, location of object mass centers, horizontal and vertical object boundaries, and object clipping. Once the objects are cut out, they are sent to a convolutional neural network, which automatically identifies the type of figure. The identification system works in real-time. The implementation was done on a Raspberry Pi 3 system and on a Jetson-Nano device. The proposal is used in an assembly course of bachelor’s degree in industrial engineering. The results presented include studying the efficiency of the recognition and processing time.

Keywords: deep-learning, image classification, image identification, industrial engineering.

Procedia PDF Downloads 134

Authors: Meruyert Serik, Nassipzhan Duisegaliyeva, Danara Tleumagambetova, Madina Ermaganbetova

Abstract:

This research paper explores the process of creating a proctoring system by evaluating the implementation of practical face recognition algorithms. Students of educational programs reviewed the research work "6B01511-Computer Science", "7M01511-Computer Science", "7M01525- STEM Education," and "8D01511-Computer Science" of Eurasian National University named after L.N. Gumilyov. As an outcome, a proctoring system will be created, enabling the conduction of tests and ensuring academic integrity checks within the system. Due to the correct operation of the system, test works are carried out. The result of the creation of the proctoring system will be the basis for the automation of the informational, educational portal developed by machine learning.

Keywords: artificial intelligence, education portal, face recognition, machine learning, proctoring

Procedia PDF Downloads 82

Authors: Alev Atak

Abstract:

In this paper, we aim to create a financial sentiment index by investigating the company’s voluntary information disclosures. We retrieve structured content from BIST 100 companies’ financial reports for the period 1998-2018 and extract relevant financial information for sentiment analysis through Natural Language Processing. We measure strategy-related disclosures and their cross-sectional variation and classify report content into generic sections using synonym lists divided into four main categories according to their liquidity risk profile, risk positions, intra-annual information, and exposure to risk. We use Word Error Rate and Cosin Similarity for comparing and measuring text similarity and derivation in sets of texts. In addition to performing text extraction, we will provide a range of text analysis options, such as the readability metrics, word counts using pre-determined lists (e.g., forward-looking, uncertainty, tone, etc.), and comparison with reference corpus (word, parts of speech and semantic level). Therefore, we create an adequate analytical tool and a financial dictionary to depict the importance of granular financial disclosure for investors to identify correctly the risk-taking behavior and hence make the aggregated effects traceable.

Keywords: financial sentiment, machine learning, information disclosure, risk

Procedia PDF Downloads 70

Authors: Ali Cheraghian, Farshid Hajati, Soheila Gheisari, Yongsheng Gao

Abstract:

In this paper, we present a novel 2.5D face recognition method based on Gabor Discrete Cosine Transform (GDCT). In the proposed method, the Gabor filter is applied to extract feature vectors from the texture and the depth information. Then, Discrete Cosine Transform (DCT) is used for dimensionality and redundancy reduction to improve computational efficiency. The system is combined texture and depth information in the decision level, which presents higher performance compared to methods, which use texture and depth information, separately. The proposed algorithm is examined on publically available Bosphorus database including models with pose variation. The experimental results show that the proposed method has a higher performance compared to the benchmark.

Keywords: Gabor filter, discrete cosine transform, 2.5d face recognition, pose

Procedia PDF Downloads 299

Authors: Nidal F. Shilbayeh, Remah W. Al-Khatib, Sameer A. Nooh

Abstract:

Arabic offline handwriting recognition systems are considered as one of the most challenging topics. Arabic Handwritten Numeral Strings are used to automate systems that deal with numbers such as postal code, banking account numbers and numbers on car plates. Segmentation of connected numerals is the main bottleneck in the handwritten numeral recognition system.  This is in turn can increase the speed and efficiency of the recognition system. In this paper, we proposed algorithms for automatic segmentation and feature extraction of Arabic handwritten numeral strings based on Watershed approach. The algorithms have been designed and implemented to achieve the main goal of segmenting and extracting the string of numeral digits written by hand especially in a courtesy amount of bank checks. The segmentation algorithm partitions the string into multiple regions that can be associated with the properties of one or more criteria. The numeral extraction algorithm extracts the numeral string digits into separated individual digit. Both algorithms for segmentation and feature extraction have been tested successfully and efficiently for all types of numerals.

Keywords: handwritten numerals, segmentation, courtesy amount, feature extraction, numeral recognition

Procedia PDF Downloads 354

Authors: Heng Yang, Tao Luo, Yakun Zhang, Kai Wang, Wei Qin, Liang Xie, Ye Yan, Erwei Yin

Abstract:

Recent years have seen that audio-visual recognition has shown great potential in a strong noise environment. The existing method of audio-visual recognition has explored methods with ResNet and feature fusion. However, on the one hand, ResNet always occupies a large amount of memory resources, restricting the application in engineering. On the other hand, the feature merging also brings some interferences in a high noise environment. In order to solve the problems, we proposed an effective framework with bidirectional distillation. At first, in consideration of the good performance in extracting of features, we chose the light model, Efficientnet as our extractor of spatial features. Secondly, self-distillation was applied to learn more information from raw data. Finally, we proposed a bidirectional distillation in decision-level fusion. In more detail, our experimental results are based on a multi-model dataset from 24 volunteers. Eventually, the lipreading accuracy of our framework was increased by 2.3% compared with existing systems, and our framework made progress in audio-visual fusion in a high noise environment compared with the system of audio recognition without visual.

Keywords: lipreading, audio-visual, Efficientnet, distillation

Procedia PDF Downloads 106

Authors: Marielle Leijten, Catherine Meulemans, Sven De Maeyer, Luuk Van Waes

Abstract:

For the diagnosis of Alzheimer's disease (AD), a large variety of neuropsychological tests are available. In some of these tests, linguistic processing - both oral and written - is an important factor. Language disturbances might serve as a strong indicator for an underlying neurodegenerative disorder like AD. However, the current diagnostic instruments for language assessment mainly focus on product measures, such as text length or number of errors, ignoring the importance of the process that leads to written or spoken language production. In this study, it is our aim to describe and test differences between cognitive and impaired elderly on the basis of a selection of writing process variables (inter- and intrapersonal characteristics). These process variables are mainly related to pause times, because the number, length, and location of pauses have proven to be an important indicator of the cognitive complexity of a process. Method: Participants that were enrolled in our research were chosen on the basis of a number of basic criteria necessary to collect reliable writing process data. Furthermore, we opted to match the thirteen cognitively impaired patients (8 MCI and 5 AD) with thirteen cognitively healthy elderly. At the start of the experiment, participants were each given a number of tests, such as the Mini-Mental State Examination test (MMSE), the Geriatric Depression Scale (GDS), the forward and backward digit span and the Edinburgh Handedness Inventory (EHI). Also, a questionnaire was used to collect socio-demographic information (age, gender, eduction) of the subjects as well as more details on their level of computer literacy. The tests and questionnaire were followed by two typing tasks and two picture description tasks. For the typing tasks participants had to copy (type) characters, words and sentences from a screen, whereas the picture description tasks each consisted of an image they had to describe in a few sentences. Both the typing and the picture description tasks were logged with Inputlog, a keystroke logging tool that allows us to log and time stamp keystroke activity to reconstruct and describe text production processes. The main rationale behind keystroke logging is that writing fluency and flow reveal traces of the underlying cognitive processes. This explains the analytical focus on pause (length, number, distribution, location, etc.) and revision (number, type, operation, embeddedness, location, etc.) characteristics. As in speech, pause times are seen as indexical of cognitive effort. Results. Preliminary analysis already showed some promising results concerning pause times before, within and after words. For all variables, mixed effects models were used that included participants as a random effect and MMSE scores, GDS scores and word categories (such as determiners and nouns) as a fixed effect. For pause times before and after words cognitively impaired patients paused longer than healthy elderly. These variables did not show an interaction effect between the group participants (cognitively impaired or healthy elderly) belonged to and word categories. However, pause times within words did show an interaction effect, which indicates pause times within certain word categories differ significantly between patients and healthy elderly.

Keywords: Alzheimer's disease, keystroke logging, matching, writing process

Procedia PDF Downloads 335

Authors: Paulo Fernando Silva Filho, Elcio Hideiti Shiguemori

Abstract:

The selection of specific landmarks for an Unmanned Aerial Vehicles’ Visual Navigation systems based on Automatic Landmark Recognition has significant influence on the precision of the system’s estimated position. At the same time, manual selection of the landmarks does not guarantee a high recognition rate, which would also result on a poor precision. This work aims to develop an automatic landmark selection that will take the image of the flight area and identify the best landmarks to be recognized by the Visual Navigation Landmark Recognition System. The criterion to select a landmark is based on features detected by ORB or AKAZE and edges information on each possible landmark. Results have shown that disposition of possible landmarks is quite different from the human perception.

Keywords: clustering, edges, feature points, landmark selection, X-means

Procedia PDF Downloads 251

Authors: Alaaeldin Hamdy Ahmed Mohammed

Abstract:

Introduction: This study examines sports’ fans and non-interested public perception and recognition of the problems that have negative impacts upon the Egyptian sports, particularly football, through caricatures. Eight caricature paintings were designed to express eight problems affecting the Egyptian sports and its development. These paintings were distributed on two groups of the fans and the non-interested public. Methods: The study was limited to eight caricatures representing the eight issues which are: the impact of stopping the sports activity on athletes, the effect of clubs’ disagreement, fanaticism between the members of the ultras of different clubs, the negative impact of the mingling of politics into sports, the negative role of the clubs affects the professionalism of the promising players, the conflict between the national organization responsible for sports, the breaking in of the fans to the playgrounds, the impact of the lack of planning on the national team. The Results: The results showed that both sports fans and those who are not interested in sports recognized the problems that the caricatures refer to and criticizes exaggeration although the rate was higher for the fans. These caricatures contributed also in their recognition of the danger of the negative impact of these problems on the Egyptian sports, particularly football which is the most common at the Egyptian sports fans. Discussion: This finding echoes the conclusion that caricatures are distinctive in the adults’ facial stimuli that are either systematically exaggerated recognition of them.

Keywords: caricature, fans, football, sports

Procedia PDF Downloads 291

Authors: Daliyah S. Aljutaili, Redna A. Almutlaq, Suha A. Alharbi, Dina M. Ibrahim

Abstract:

All currencies around the world look very different from each other. For instance, the size, color, and pattern of the paper are different. With the development of modern banking services, automatic methods for paper currency recognition become important in many applications like vending machines. One of the currency recognition architecture’s phases is Feature detection and description. There are many algorithms that are used for this phase, but they still have some disadvantages. This paper proposes a feature detection algorithm, which merges the advantages given in the current SIFT and SURF algorithms, which we call, Speeded up Robust Scale-Invariant Feature Transform (SR-SIFT) algorithm. Our proposed SR-SIFT algorithm overcomes the problems of both the SIFT and SURF algorithms. The proposed algorithm aims to speed up the SIFT feature detection algorithm and keep it robust. Simulation results demonstrate that the proposed SR-SIFT algorithm decreases the average response time, especially in small and minimum number of best key points, increases the distribution of the number of best key points on the surface of the currency. Furthermore, the proposed algorithm increases the accuracy of the true best point distribution inside the currency edge than the other two algorithms.

Keywords: currency recognition, feature detection and description, SIFT algorithm, SURF algorithm, speeded up and robust features

Procedia PDF Downloads 212

Authors: Bowen Ding, Yihao Kuang

Abstract:

In recent years, large-scale pre-trained language models have achieved state-of-the-art performance on a variety of natural language processing tasks. The word vectors produced by these language models can be viewed as dense encoded presentations of natural language that in text form. However, it is unknown how much linguistic information is encoded and how. In this paper, we construct several corresponding probing tasks for multiple linguistic information to clarify the encoding capabilities of different language models and performed a visual display. We firstly obtain word presentations in vector form from different language models, including BERT, ELMo, RoBERTa and GPT. Classifiers with a small scale of parameters and unsupervised tasks are then applied on these word vectors to discriminate their capability to encode corresponding linguistic information. The constructed probe tasks contain both semantic and syntactic aspects. The semantic aspect includes the ability of the model to understand semantic entities such as numbers, time, and characters, and the grammatical aspect includes the ability of the language model to understand grammatical structures such as dependency relationships and reference relationships. We also compare encoding capabilities of different layers in the same language model to infer how linguistic information is encoded in the model.

Keywords: language models, probing task, text presentation, linguistic information

Procedia PDF Downloads 75

Authors: Saeedeh Barzegar Khalilsaraei, Manoocheher Kelarestaghi, Farshad Eshghi

Abstract:

The vehicle color, as a prominent and stable feature, helps to identify a vehicle more accurately. As a result, vehicle color recognition is of great importance in intelligent transportation systems. Unlike conventional methods which use only a single Convolutional Neural Network (CNN) for feature extraction or classification, in this paper, four CNNs, with different architectures well-performing in different classes, are trained to extract various features from the input image. To take advantage of the distinct capability of each network, the multiple outputs are combined using a stack generalization algorithm as an ensemble technique. As a result, the final model performs better than each CNN individually in vehicle color identification. The evaluation results in terms of overall average accuracy and accuracy variance show the proposed method’s outperformance compared to the state-of-the-art rivals.

Keywords: Vehicle Color Recognition, Ensemble Algorithm, Stack Generalization, Convolutional Neural Network

Procedia PDF Downloads 53

Authors: Madhavi

Abstract:

All linguistic units are context-dependent. They occur in particular settings, from which they derive much of their import, and are recognized by speakers as distinct entities only through a process of abstraction. Most of the words have several concepts associated with them and convey a number of meanings in different contexts in any language. For instance, there are different uses of the word good as an adjective from English. The adjective good expresses many senses like (1) ‘high quality of someone or something’ (2) ‘efficient’ (3) ‘virtuous’ (4) ‘reliable’ etc. These senses will be analyzed by using cognitive semantics framework. The context has the power to insulate one meaning from all the other meanings in communication. This paper will provide a cognitive semantic analysis. The basic tenet of cognitive semantics is the sense of a word is the way we conceptualize it. Our conceptualization is based on the physical experience we go through. Cognitive semantics tries to capture this conceptualization in terms of some categories like schema, frame, and domain. Cognitive semantics is a subfield of cognitive linguistics. Cognitive linguistics studies the language creation, learning, and usage by the reference to human cognition. The semantic structure is conceptual structure which is related to the concepts which are the elements of reason and constitute the meanings of words and linguistic expressions. Cognitive semantics studies how our mind works for the meaning of any word and how it perceives meaning from the environment through senses and works to map with the knowledge which already exists in our mind through experience. In the present paper, the senses are further classified into some categories.

Keywords: cognitive, contexts, semantics, senses

Procedia PDF Downloads 193

Authors: Nawab Yousuf Ali, S. Golam, A. Ameer, Ashok Toru Roy

Abstract:

This paper develops new morphological rules suitable for Bangla repetition words to be incorporated into an inter lingua representation called Universal Networking Language (UNL). The proposed rules are to be used to combine verb roots and their inflexions to produce words which are then combined with other similar types of words to generate repetition words. This paper outlines the format of morphological rules for different types of repetition words that come from verb roots based on the framework of UNL provided by the UNL centre of the Universal Networking Digital Language (UNDL) foundation.

Keywords: Universal Networking Language (UNL), universal word (UW), head word (HW), Bangla-UNL Dictionary, morphological rule, enconverter (EnCo)

Procedia PDF Downloads 285

Authors: Lanchi Xie, Zhihui Li, Zhigang Li, Guiqiang Wang, Lei Xu, Yuwen Yan

Abstract:

Image-based facial features can be classified into category recognition features and individual recognition features. Current automated face recognition systems extract a specific feature vector of different dimensions from a facial image according to their pre-trained neural network. However, to improve the efficiency of parameter calculation, an algorithm generally reduces the image details by pooling. The operation will overlook the details concerned much by forensic experts. In our experiment, we adopted a variety of face recognition algorithms based on deep learning, compared a large number of naturally collected face images with the known data of the same person's frontal ID photos. Downscaling and manual handling were performed on the testing images. The results supported that the facial recognition algorithms based on deep learning detected structural and morphological information and rarely focused on specific markers such as stains and moles. Overall performance, distribution of genuine scores and impostor scores, and likelihood ratios were tested to evaluate the accuracy of biometric systems and forensic experts. Experiments showed that the biometric systems were skilled in distinguishing category features, and forensic experts were better at discovering the individual features of human faces. In the proposed approach, a fusion was performed at the score level. At the specified false accept rate, the framework achieved a lower false reject rate. This paper contributes to improving the interpretability of the objective method of facial comparison and provides a novel method for human-machine collaboration in this field.

Keywords: likelihood ratio, automated facial recognition, facial comparison, biometrics

Procedia PDF Downloads 102

Authors: Jun Gil Ahn, Jong Kang Park, Jong Tae Kim

Abstract:

In this paper, we analyze major components of activity recognition (AR) in wearable device with 9-axis sensors and sensor fusion filters. 9-axis sensors commonly include 3-axis accelerometer, 3-axis gyroscope and 3-axis magnetometer. We chose sensor fusion filters as Kalman filter and Direction Cosine Matrix (DCM) filter. We also construct sensor fusion data from each activity sensor data and perform classification by accuracy of AR using Naïve Bayes and SVM. According to the classification results, we observed that the DCM filter and the specific combination of the sensing axes are more effective for AR in wearable devices while classifying walking, running, ascending and descending.

Keywords: accelerometer, activity recognition, directiona cosine matrix filter, gyroscope, Kalman filter, magnetometer

Procedia PDF Downloads 300

Authors: Koray U. Erbas

Abstract:

Neural networks are appealing for many applications since they are able to learn complex non-linear relationships between input and output data. As the number of neurons and layers in a neural network increase, it is possible to represent more complex relationships with automatically extracted features. Nowadays Deep Neural Networks (DNNs) are widely used in Computer Vision problems such as; classification, object detection, segmentation image editing etc. In this work, Facial Emotion Recognition task is performed by proposed Convolutional Neural Network (CNN)-based DNN architecture using FER2013 Dataset. Moreover, the effects of different hyperparameters (activation function, kernel size, initializer, batch size and network size) are investigated and ablation study results for Pooling Layer, Dropout and Batch Normalization are presented.

Keywords: convolutional neural network, deep learning, deep learning based FER, facial emotion recognition

Procedia PDF Downloads 233

Authors: Carlos Vera, Tetyana Baydyk, Ernst Kussul, Graciela Velasco, Miguel Aparicio

Abstract:

This article describes the Random Subspace Neural Classifier (RSC) for the recognition of meteors in the night sky. We used images of meteors entering the atmosphere at night between 8:00 p.m.-5: 00 a.m. The objective of this project is to classify meteor and star images (with stars as the image background). The monitoring of the sky and the classification of meteors are made for future applications by scientists. The image database was collected from different websites. We worked with RGB-type images with dimensions of 220x220 pixels stored in the BitMap Protocol (BMP) format. Subsequent window scanning and processing were carried out for each image. The scan window where the characteristics were extracted had the size of 20x20 pixels with a scanning step size of 10 pixels. Brightness, contrast and contour orientation histograms were used as inputs for the RSC. The RSC worked with two classes and classified into: 1) with meteors and 2) without meteors. Different tests were carried out by varying the number of training cycles and the number of images for training and recognition. The percentage error for the neural classifier was calculated. The results show a good RSC classifier response with 89% correct recognition. The results of these experiments are presented and discussed.

Keywords: contour orientation histogram, meteors, night sky, RSC neural classifier, stars

Procedia PDF Downloads 121

Authors: Wanhyun Cho, Soonja Kang, Sangkyoon Kim, Soonyoung Park

Abstract:

In this paper, we present the human action recognition method using the variational Bayesian HMM with the Dirichlet process mixture (DPM) of the Gaussian-Wishart emission model (GWEM). First, we define the Bayesian HMM based on the Dirichlet process, which allows an infinite number of Gaussian-Wishart components to support continuous emission observations. Second, we have considered an efficient variational Bayesian inference method that can be applied to drive the posterior distribution of hidden variables and model parameters for the proposed model based on training data. And then we have derived the predictive distribution that may be used to classify new action. Third, the paper proposes a process of extracting appropriate spatial-temporal feature vectors that can be used to recognize a wide range of human behaviors from input video image. Finally, we have conducted experiments that can evaluate the performance of the proposed method. The experimental results show that the method presented is more efficient with human action recognition than existing methods.

Keywords: human action recognition, Bayesian HMM, Dirichlet process mixture model, Gaussian-Wishart emission model, Variational Bayesian inference, prior distribution and approximate posterior distribution, KTH dataset

Procedia PDF Downloads 322

Authors: B. R. Campomanes-Alvarez, P. Quiros, B. Fernandez

Abstract:

Automatic Speech Recognition (ASR) is a machine-based process of decoding and transcribing oral speech. A typical ASR system receives acoustic input from a speaker or an audio file, analyzes it using algorithms, and produces an output in the form of a text. Some speech recognition systems use Hidden Markov Models (HMMs) to deal with the temporal variability of speech and Gaussian Mixture Models (GMMs) to determine how well each state of each HMM fits a short window of frames of coefficients that represents the acoustic input. Another way to evaluate the fit is to use a feed-forward neural network that takes several frames of coefficients as input and produces posterior probabilities over HMM states as output. Deep neural networks (DNNs) that have many hidden layers and are trained using new methods have been shown to outperform GMMs on a variety of speech recognition systems. Acoustic models for state-of-the-art ASR systems are usually training on massive amounts of data. However, audio files with their corresponding transcriptions can be difficult to obtain, especially in the Spanish language. Hence, in the case of these low-resource scenarios, building an ASR model is considered as a complex task due to the lack of labeled data, resulting in an under-trained system. Semi-supervised learning approaches arise as necessary tasks given the high cost of transcribing audio data. The main goal of this proposal is to develop a procedure based on acoustic semi-supervised learning for Spanish ASR systems by using DNNs. This semi-supervised learning approach consists of: (a) Training a seed ASR model with a DNN using a set of audios and their respective transcriptions. A DNN with a one-hidden-layer network was initialized; increasing the number of hidden layers in training, to a five. A refinement, which consisted of the weight matrix plus bias term and a Stochastic Gradient Descent (SGD) training were also performed. The objective function was the cross-entropy criterion. (b) Decoding/testing a set of unlabeled data with the obtained seed model. (c) Selecting a suitable subset of the validated data to retrain the seed model, thereby improving its performance on the target test set. To choose the most precise transcriptions, three confidence scores or metrics, regarding the lattice concept (based on the graph cost, the acoustic cost and a combination of both), was performed as selection technique. The performance of the ASR system will be calculated by means of the Word Error Rate (WER). The test dataset was renewed in order to extract the new transcriptions added to the training dataset. Some experiments were carried out in order to select the best ASR results. A comparison between a GMM-based model without retraining and the DNN proposed system was also made under the same conditions. Results showed that the semi-supervised ASR-model based on DNNs outperformed the GMM-model, in terms of WER, in all tested cases. The best result obtained an improvement of 6% relative WER. Hence, these promising results suggest that the proposed technique could be suitable for building ASR models in low-resource environments.

Keywords: automatic speech recognition, deep neural networks, machine learning, semi-supervised learning

Procedia PDF Downloads 320