J. Amudha

Vehicle Detection in Static Images Using Color and Corner Map

This paper presents an approach to identify the vehicle in the static images using color and corner map. The detection of vehicles in a traffic scene can address wide range of traffic problems. Here an attempt has been made to reduce the search time to find the possible vehicle candidates thereby reducing the computation time without a full search. A color transformation is used to project all the colors of input pixels to a new feature space such that vehicle pixels can be easily distinguished from non-vehicle ones. Bayesian classifier is adopted for verifying the vehicle pixels from the background. Corner map is used for removing the false detections and to verify the vehicle candidates.

Mid-Point Hough Transform: A Fast Line Detection Method

This paper proposes a new method for detection of lines in images. The new algorithm is a modification of the Standard Hough Transform by considering a pair of pixels simultaneously and mapping them to the parameter space. The proposed algorithm is compared with line detection algorithms like Standard Hough Transform, Randomized Hough Transform and its variants and has advantages of smaller storage and higher speed. I. INTRODUCTION Detection of lines in a image is one of the basic task in image processing. It has got several applications in a number of fields like sonars, mobile applications and biometric security systems. In sonar systems the presence of a target and its

Classification and feature vector techniques to improve fractal image coding

Fractal image compression receives much attention because of its desirable properties like resolution independence, fast decoding and very competitive rate-distortion curves. Despite the advances made in fractal image compression the long computing time in encoding phase still remain as main drawback of this technique as encoding step is computationally expensive. A large number of sequential searches through portions of the image are carried out to identify best matches for other image portions. So far, several methods have been proposed in order to speed-up fractal image coding. Here an attempt is made to analyze the speed-up techniques like classification and feature vector, which demonstrates the search through larger portions of the domain pool without increasing computation time. In this way both the image quality and compression ratio are improved at reduced computation time. Experimental results and analysis show that proposed method can speed up fractal image encoding process over conventional methods.

Feature Selection in Top-Down Visual Attention Model using WEKA.

A feature selection in Top down visual attention model for sign board recognition has been incorporated to reduce the computational complexity and to enhance the quality of recognition. The approach is based on a biologically motivated attention system which is able to detect regions of interest in images based on the concepts of the human visual system. A top-down guided visual search module of the system identifies the most discriminate feature from the previously learned target object and uses to recognize the object. This enables a significantly faster classification and is illustrated in identifying signboards in a road scene environment. General Terms Computer Vision, Pattern Recognition, Biological Vision

Suitability of Genetic Algorithm and Particle Swarm Optimization for Eye Tracking System

Evolutionary algorithms provide solutions to optimization problem and its suitability to eye tracking is explored in this paper. In this paper, we compare the evolutionary methods Genetic Algorithm (GA) and Particle Swarm Optimization (PSO) using deformable template matching for eye tracking. Here we address the various eye tracking challenges like head movements, eye movements, eye blinking and zooming that affect the efficiency of the system. GA and PSO based Eye tracking systems are presented for real time video sequence. Eye detection is done by Haar-like features. For eye tracking, GAET and PSOET use deformable template matching to find the best solution. The experimental results show that PSOET achieves tracking accuracy of 98% in less time. GAET predicted eye has high correlation to actual eye but the tracking accuracy is only 91 %.

Developing an application using eye tracker

Eye tracking measures where the eye is focused or the movement of eye with respect to the head. The eye tracker will track the eye positions and eye movement for the visual stimulus presented on the computer system. Various features like gaze point, pupil size and mouse position can be extracted and it can be represented using visualization techniques such as fixation, saccade, scanpath and heat map. The features obtained from eye tracker can be extended to real life applications. Using this technology companies could be able to analyze thousands of customers eye patterns in real-time, and make decisions on marketing based on the data. The technology can analyze the stress level of patients, employees in IT, BPO, accounting, banking, front office etc. Here we are illustrating the advantages and applications of eye tracking, its usability and how to develop an application using a commercial eye tracker.

An Android-Based Mobile Eye Gaze Point Estimation System for Studying the Visual Perception in Children with Autism

Autism is a neural developmental disorder characterized by poor social interaction, communication impairments and repeated behaviour. Reason for this difference in behaviour can be understood by studying the difference in their sensory processing. This paper proposes a mobile application which uses visual tasks to study the visual perception in children with Autism, which can give a profound explanation to the fact why they see and perceive things differently when compared to normal children. The application records the eye movements and estimates the region of gaze of the child to understand where the child’s attention focuses to during the visual tasks. This work provides an experimental proof that children with Autism are superior when compared to normal children in some visual tasks, which proves that they have higher IQ levels than their peers.

Comparative Study of Visual Attention Models with Human Eye Gaze in Remote Sensing Images

Computational visual attention model analogous to human eye behaviour has a tremendous need for applications in various fields. Study of visual attention yields information about a persons conscious processing while performing a task. This paper evaluates the behaviour of the bottom-up visual attention models with the eye gaze data set based on various performance metrics. The eye tracking data used for the study measures the gaze fixation points of human beings viewing remote sensing images. The evaluation of the models concludes that the Graph Based Visual saliency model predicts better than the Itti-Koch model across all the performance measures such as Normalized Scanpath Saliency (NSS) score, Area Under the Curve (AUC) score and Linear Correlation Coefficient for the remote sensing images.

Saliency based modified chamfers matching method for sketch based image retrieval

Recent advances of tablet PC and multi-touch screen technology raised increasing interest on users search and retrieve the desired images in databases in a simple manner. Sketch based image retrieval (SBIR) emerged as a more expressive and interactive way to perform image search. Our focuses on this work is to enhance the methods of Indexable Oriented Chamfer Matching using salient feature detection algorithms. As per psychologist observation, edge is one of the most dominant feature of an image to represent the content. Prioritization is done based on edge content of image by ignoring parameters like color, texture. Experiments are done in benchmark dataset of [11] demonstrate the better performance of our approach.

A Generic Bio-inspired Framework for Detecting Humans Based on Saliency Detection

Even with all its advancement in technology, computer vision system cannot competes with nature’s gift—the brains, that arranges the objects quickly and extract the necessary information from huge data. A bio-inspired feature selection method is proposed for detecting the humans using saliency detection. It is performed by tuning prominent features such as color, orientation, and intensity in bottom-up approach to locate the probable candidate regions of humans in an image. Further, the results improved in detection phase that makes use of weights learned from training samples to ignore non-human regions in the candidate regions. The overall system has an accuracy rate of 90 % for detecting the human region.