Search results for: decoder

Authors: Mohd Ayyub Khan, Nadeem Akhtar

Abstract:

Due to recent advances in wireless communication technologies and hand-held devices, there is a huge demand for video-based applications such as video surveillance, video conferencing, remote surgery, Digital Video Broadcast (DVB), IPTV, online learning courses, YouTube, WhatsApp, Instagram, Facebook, Interactive Video Games. However, the raw videos posses very high bandwidth which makes the compression a must before its transmission over the wireless channels. The High Efficiency Video Codec (HEVC) (also called H.265) is latest state-of-the-art video coding standard developed by the Joint effort of ITU-T and ISO/IEC teams. HEVC is targeted for high resolution videos such as 4K or 8K resolutions that can fulfil the recent demands for video services. The compression ratio achieved by the HEVC is twice as compared to its predecessor H.264/AVC for same quality level. The compression efficiency is generally increased by removing more correlation between the frames/pixels using complex techniques such as extensive intra and inter prediction techniques. As more correlation is removed, the chances of interdependency among coded bits increases. Thus, bit errors may have large effect on the reconstructed video. Sometimes even single bit error can lead to catastrophic failure of the reconstructed video. In this paper, we study the performance of HEVC bitstream over additive white Gaussian noise (AWGN) channel. Moreover, HEVC over Quadrature Amplitude Modulation (QAM) combined with forward error correction (FEC) schemes are also explored over the noisy channel. The video will be encoded using HEVC, and the coded bitstream is channel coded to provide some redundancies. The channel coded bitstream is then modulated using QAM and transmitted over AWGN channel. At the receiver, the symbols are demodulated and channel decoded to obtain the video bitstream. The bitstream is then used to reconstruct the video using HEVC decoder. It is observed that as the signal to noise ratio of channel is decreased the quality of the reconstructed video decreases drastically. Using proper FEC codes, the quality of the video can be restored up to certain extent. Thus, the performance analysis of HEVC presented in this paper may assist in designing the optimized code rate of FEC such that the quality of the reconstructed video is maximized over wireless channels.

Keywords: AWGN, forward error correction, HEVC, video coding, QAM

Procedia PDF Downloads 123

Authors: Paola Causin, Andrea Aspri, Alessandro Benfenati

Abstract:

Diffuse Optical Tomography (DOT) is an emergent medical imaging technique which employs NIR light to estimate the spatial distribution of optical coefficients in biological tissues for diagnostic purposes, in a noninvasive and non-ionizing manner. DOT reconstruction is a severely ill-conditioned problem due to prevalent scattering of light in the tissue. In this contribution, we present our research in adopting hybrid knowledgedriven/data-driven approaches which exploit the existence of well assessed physical models and build upon them neural networks integrating the availability of data. Namely, since in this context regularization procedures are mandatory to obtain a reasonable reconstruction [1], we explore the use of neural networks as tools to include prior information on the solution. 2. Materials and Methods The idea underlying our approach is to leverage neural networks to solve PDE-constrained inverse problems of the form 𝒒 ∗ = 𝒂𝒓𝒈 𝒎𝒊𝒏𝒒 𝐃(𝒚, 𝒚̃), (1) where D is a loss function which typically contains a discrepancy measure (or data fidelity) term plus other possible ad-hoc designed terms enforcing specific constraints. In the context of inverse problems like (1), one seeks the optimal set of physical parameters q, given the set of observations y. Moreover, 𝑦̃ is the computable approximation of y, which may be as well obtained from a neural network but also in a classic way via the resolution of a PDE with given input coefficients (forward problem, Fig.1 box ). Due to the severe ill conditioning of the reconstruction problem, we adopt a two-fold approach: i) we restrict the solutions (optical coefficients) to lie in a lower-dimensional subspace generated by auto-decoder type networks. This procedure forms priors of the solution (Fig.1 box ); ii) we use regularization procedures of type 𝒒̂ ∗ = 𝒂𝒓𝒈𝒎𝒊𝒏𝒒 𝐃(𝒚, 𝒚̃)+ 𝑹(𝒒), where 𝑹(𝒒) is a regularization functional depending on regularization parameters which can be fixed a-priori or learned via a neural network in a data-driven modality. To further improve the generalizability of the proposed framework, we also infuse physics knowledge via soft penalty constraints (Fig.1 box ) in the overall optimization procedure (Fig.1 box ). 3. Discussion and Conclusion DOT reconstruction is severely hindered by ill-conditioning. The combined use of data-driven and knowledgedriven elements is beneficial and allows to obtain improved results, especially with a restricted dataset and in presence of variable sources of noise.

Keywords: inverse problem in tomography, deep learning, diffuse optical tomography, regularization

Procedia PDF Downloads 44

Authors: Shubham Gogri, Lucia Florescu

Abstract:

Medical diagnosis and radiotherapy treatment planning using Computed Tomography (CT) rely on the quantitative accuracy and quality of the CT images. At the same time, requirements for CT imaging include reducing the radiation dose exposure to patients and minimizing scanning time. A solution to this is the sparse-view CT technique, based on a reduced number of projection views. This, however, introduces a new problem— the incomplete projection data results in lower quality of the reconstructed images. To tackle this issue, deep learning methods have been applied to enhance the quality of the sparse-view CT images. A first approach involved employing Mir-Net, a dedicated deep neural network designed for image enhancement. This showed promise, utilizing an intricate architecture comprising encoder and decoder networks, along with the incorporation of the Charbonnier Loss. However, this approach was computationally demanding. Subsequently, a specialized Generative Adversarial Network (GAN) architecture, rooted in the Pix2Pix framework, was implemented. This GAN framework involves a U-Net-based Generator and a Discriminator based on Convolutional Neural Networks. To bolster the GAN's performance, both Charbonnier and Wasserstein loss functions were introduced, collectively focusing on capturing minute details while ensuring training stability. The integration of the perceptual loss, calculated based on feature vectors extracted from the VGG16 network pretrained on the ImageNet dataset, further enhanced the network's ability to synthesize relevant images. A series of comprehensive experiments with clinical CT data were conducted, exploring various GAN loss functions, including Wasserstein, Charbonnier, and perceptual loss. The outcomes demonstrated significant image quality improvements, confirmed through pertinent metrics such as Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index (SSIM) between the corrected images and the ground truth. Furthermore, learning curves and qualitative comparisons added evidence of the enhanced image quality and the network's increased stability, while preserving pixel value intensity. The experiments underscored the potential of deep learning frameworks in enhancing the visual interpretation of CT scans, achieving outcomes with SSIM values close to one and PSNR values reaching up to 76.

Keywords: generative adversarial networks, sparse view computed tomography, CT image correction, Mir-Net

Procedia PDF Downloads 117

Authors: Szymon Rutkowski, Tomasz Korbak

Abstract:

We present a reinforcement-learning-based method of training neural machine translation models without parallel corpora. The standard encoder-decoder approach to machine translation suffers from two problems we aim to address. First, it needs parallel corpora, which are scarce, especially for low-resource languages. Second, it lacks psychological plausibility of learning procedure: learning a foreign language is about learning to communicate useful information, not merely learning to transduce from one language’s 'encoding' to another. We instead pose the problem of learning to translate as learning a policy in a communication game between two agents: the translator and the classifier. The classifier is trained beforehand on a natural language inference task (determining the entailment relation between a premise and a hypothesis) in the target language. The translator produces a sequence of actions that correspond to generating translations of both the hypothesis and premise, which are then passed to the classifier. The translator is rewarded for classifier’s performance on determining entailment between sentences translated by the translator to disciple’s native language. Translator’s performance thus reflects its ability to communicate useful information to the classifier. In effect, we train a machine translation model without the need for parallel corpora altogether. While similar reinforcement learning formulations for zero-shot translation were proposed before, there is a number of improvements we introduce. While prior research aimed at grounding the translation task in the physical world by evaluating agents on an image captioning task, we found that using a linguistic task is more sample-efficient. Natural language inference (also known as recognizing textual entailment) captures semantic properties of sentence pairs that are poorly correlated with semantic similarity, thus enforcing basic understanding of the role played by compositionality. It has been shown that models trained recognizing textual entailment produce high-quality general-purpose sentence embeddings transferrable to other tasks. We use stanford natural language inference (SNLI) dataset as well as its analogous datasets for French (XNLI) and Polish (CDSCorpus). Textual entailment corpora can be obtained relatively easily for any language, which makes our approach more extensible to low-resource languages than traditional approaches based on parallel corpora. We evaluated a number of reinforcement learning algorithms (including policy gradients and actor-critic) to solve the problem of translator’s policy optimization and found that our attempts yield some promising improvements over previous approaches to reinforcement-learning based zero-shot machine translation.

Keywords: agent-based language learning, low-resource translation, natural language inference, neural machine translation, reinforcement learning

Procedia PDF Downloads 98

Authors: Anjalika Fernando, Banuka Athuraliya

Abstract:

This paper presents a comprehensive approach for converting Sinhala Sign Language (SSL) into grammatically correct sentences using Natural Language Processing (NLP) techniques in real-time. While previous studies have explored various aspects of SSL translation, the research gap lies in the absence of grammar checking for SSL. This work aims to bridge this gap by proposing a two-stage methodology that leverages deep learning models to detect signs and translate them into coherent sentences, ensuring grammatical accuracy. The first stage of the approach involves the utilization of a Long Short-Term Memory (LSTM) deep learning model to recognize and interpret SSL signs. By training the LSTM model on a dataset of SSL gestures, it learns to accurately classify and translate these signs into textual representations. The LSTM model achieves a commendable accuracy rate of 94%, demonstrating its effectiveness in accurately recognizing and translating SSL gestures. Building upon the successful recognition and translation of SSL signs, the second stage of the methodology focuses on improving the grammatical correctness of the translated sentences. The project employs a Neural Machine Translation (NMT) architecture, consisting of an encoder and decoder with LSTM components, to enhance the syntactical structure of the generated sentences. By training the NMT model on a parallel corpus of Sinhala wrong sentences and their corresponding grammatically correct translations, it learns to generate coherent and grammatically accurate sentences. The NMT model achieves an impressive accuracy rate of 98%, affirming its capability to produce linguistically sound translations. The proposed approach offers significant contributions to the field of SSL translation and grammar correction. Addressing the critical issue of grammar checking, it enhances the usability and reliability of SSL translation systems, facilitating effective communication between hearing-impaired and non-sign language users. Furthermore, the integration of deep learning techniques, such as LSTM and NMT, ensures the accuracy and robustness of the translation process. This research holds great potential for practical applications, including educational platforms, accessibility tools, and communication aids for the hearing-impaired. Furthermore, it lays the foundation for future advancements in SSL translation systems, fostering inclusive and equal opportunities for the deaf community. Future work includes expanding the existing datasets to further improve the accuracy and generalization of the SSL translation system. Additionally, the development of a dedicated mobile application would enhance the accessibility and convenience of SSL translation on handheld devices. Furthermore, efforts will be made to enhance the current application for educational purposes, enabling individuals to learn and practice SSL more effectively. Another area of future exploration involves enabling two-way communication, allowing seamless interaction between sign-language users and non-sign-language users.In conclusion, this paper presents a novel approach for converting Sinhala Sign Language gestures into grammatically correct sentences using NLP techniques in real time. The two-stage methodology, comprising an LSTM model for sign detection and translation and an NMT model for grammar correction, achieves high accuracy rates of 94% and 98%, respectively. By addressing the lack of grammar checking in existing SSL translation research, this work contributes significantly to the development of more accurate and reliable SSL translation systems, thereby fostering effective communication and inclusivity for the hearing-impaired community

Keywords: Sinhala sign language, sign Language, NLP, LSTM, NMT

Procedia PDF Downloads 71