Search

Bilateral filter is an edge-preserving smoother that has applications in image processing, computer vision, and computational photography. In the past, field-programmable gate array (FPGA) implementations of the filter have been proposed that can achieve high throughput using parallelization and pipelining. An inherent limitation with direct implementations is that their complexity scales as $O(\omega ^2)$ with the filter width $\omega$ . In this paper, we propose an FPGA implementation of a fast bilateral filter that requires just $O(1)$ operations for any arbitrary $\omega$ . The attractive feature of the FPGA implementation is that it is both scalable and reconfigurable. To the best of our knowledge, this is the first scalable FPGA implementation of the bilateral filter. As an application, we use the FPGA implementation for image denoising.

Deaf or hard-of-hearing people mostly rely on lip-reading to understand speech. They demonstrate the ability of humans to understand speech from visual cues only. Automatic lip reading systems work in a similar fashion - by obtaining speech or text from just the visual information, like a video of a person's face. In this paper, an automatic lip reading system for spoken digit recognition is presented. The system uses simple dynamic features by creating difference images between consecutive frames of the video input. Using this technique, word recognition rates of 83.79 and 65.58 are achieved in speaker-dependent and speaker-independent testing scenarios, respectively. A novel, extended region-of-interest (ROI) which includes lower jaw and neck region is also introduced. Most lip-reading algorithms use only the mouth/lip region for relevant feature extraction. Over simple mouth as the ROI, the proposed ROI improves the performance by 4 in speaker-dependent tests and by 11 in speaker-independent tests. © 2018 Association for Computing Machinery.

Visual Speech Recognition (VSR) deals with the task of extracting speech information from visual cues from a person's face while speaking. Accurate lip segmentation and modeling are essential in any VSR algorithm for good feature extraction. However, lip modeling is a complicated task and is not very robust in natural conditions. This paper describes a novel technique for limited vocabulary visual-only speech recognition that does not use lip modeling. For visual feature extraction, Discrete Cosine Transform (DCT) and Local Binary Pattern (LBP) have been tested. An Error-Correcting Output Codes (ECOC) multi-class model using Support Vector Machine (SVM) binary learners is used for recognition and classification of words.

Background/Objectives: Speech emotion recognition (SER) is an important aspect of Human-Computer Interaction systems which is widely used in different sectors like healthcare, robotics, automatic call centres and distance education. Speech emotion recognition involves in depth analysis of the signal and identifying the appropriate emotion based on its trained database using extracted features. Method/Statistical Analysis: This paper aims in devising SER system using linear prediction of the causal part of the autocorrelation sequence (OSALPC) algorithm which has been proven to efficiently reduce noise along with Linear Frequency Cepstral Coefficients (LFCC), Linear Predictive Coding (LPC), MFCC, LPC using cepstrum for feature extraction. After extracting the feature vectors from the voice signal, it is modelled using Gaussian Mixture Models (GMM). The MAP (Maximum a posteriori) rule is used for decision making. Findings: Performance was analysed and our proposed system showed an overall efficiency of 89% when tested on German database (Emo-DB) for 7 emotions. The overall efficiency has proven to increase compared to the studies made up to date on the German Database. The highest emotion recognition rate was for SAD using fused algorithm which was 95.56%. Also results were tabulated and compared using Modified MFCC. A Graphical unit interface of the proposed system is also devised. Application/Improvements: The applications of speech emotion recognition are farfetched. Further scope of this work will be a comparison of the achieved recognition rate using algorithms with recognition rate achieved by humans.

Enhanced Local Ternary Patterns (ELTP) significantly improves performance over other feature descriptor methods including Local Binary Patterns (LBP) and Local Ternary Patterns (LTP). Sequential implementation of ELTP results in poor performance in terms of execution time for real time systems. Speed and accuracy are important characteristics of a real time face recognition system. With the aim of fulfilling both these criteria, this paper presents an implementation of GPU Accelerated Face Recognition System with ELTP using OpenCL framework. As a result of our Optimization techniques, we have achieved highest kernel execution speedup of 374x for ELTP image and histogram generation with 4096 x H 4096 (16MP) image resolution when it is implemented on GPU. Face recognition with ELTP showed higher recognition rates on ORL database. We also implemented LBP and LTP algorithms on GPU and compared their performances with ELTP. Similar Optimization techniques were applied for LBP kernel executions, which resulted in much higher speedups when compared to their previous implementations. Experimental results demonstrated that Parallel implementation with ELTP on GPU (AMD Radeon HD 7650M) outperforms CPU based face recognition system using LBP in terms of speed and accuracy.

Distributed processing is a new paradigm to analyse the huge volume of video data in camera networks. This paper addresses the problem of distributed single target tracking considering false positives and missed detections. Target tracking is modelled as a dynamic state estimation problem with nonlinear process and measurement model. We propose to use the sigma point information filters combined with a consensus algorithm. Sigma point information filters are integrated with probabilistic data association filter to deal with false positives and missed detections. We use a distributed average consensus algorithm which converges in finite time. Unlike other related state of the art technique papers, we report results on real data and show the effectiveness of the proposed algorithm.

This paper presents a real-time face recognition system using kinect sensor. The algorithm is implemented on GPU using opencl and significant speed improvements are observed. We use kinect depth image to increase the robustness and reduce computational cost of conventional LBP based face recognition. The main objective of this paper was to perform robust, high speed fusion based face recognition and tracking. The algorithm is mainly composed of three steps. First step is to detect all faces in the video using viola jones algorithm. The second step is online database generation using a tracking window on the face. A modified LBP feature vector is calculated using fusion information from depth and greyscale image on GPU. This feature vector is used to train a svm classifier. Third step involves recognition of multiple faces based on our modified feature vector.

This paper presents a novel approach for real time face detection using heterogeneous computing. The algorithm uses local binary pattern (LBP) as feature vector for face detection. OpenCL is used to accelerate the code using GPU[1]. Illuminance invariance is achieved using gamma correction and Difference of Gaussian(DOG) to make the algorithm robust against varying lighting conditions. This implementation is compared with previous parallel implementation and is found to perform faster.

With the advent of Internet, video over IP is gaining popularity. In such an enviroment, scalability and fault tolerance will be the key issues. Existing video on demand (VoD) service systems [1], [2], [3] are usually neither scalable nor tolerant to server faults and hence fail to comply to multi-user, failure-prone networks such as the Internet. Current research areas concerning VoD often focus on increasing the throughput and reliability of single server, but rarely addresses the smooth provision of service during server as well as network failures. Reliable Server Pooling (RSerPool), being capable of providing high availability by using multiple redundant servers as single source point, can be a solution to overcome the above failures. During a possible server failure, the continuity of service is retained by another server. In order to achieve transparent failover, efficient state sharing is an important requirement. In this paper, we present an elegant, simple, efficient and scalable approach which has been developed to facilitate the transfer of state by the client itself, using extended cookie mechanism, which ensures that there is no noticeable change in disruption or the video quality.