Design and Performance Analysis of Sign Language Recognition System using Hybrid Feature Descriptor

Abstract

In recent times, images and videos have emerged as one of essential information sources newlinedepicting real-time scenarios. Digital images nowadays serve as input for many newlineapplications and replace the manual methods due to their capabilities of 3D scene newlinerepresentation in a 2D plane. The capabilities of digital images and the utilization of newlinemachine learning methodologies show good accuracy in many prediction and pattern newlinerecognition applications. One of the application fields pertains to sign language newlinerecognition, which acts as a communication tool for sensory impaired persons. newlineTraditionally interpreter performs all the tasks related to the hand sign images newlineinterpretation. But the presence of an interpreter for every sensory impaired is not possible newlineand does not suffice the accuracy levels. This creates room for research in developing newlineautomation-based methods where the images captured through sensors and cameras will newlinebe used for the recognition of sign language. The digital images captured with the help of newlinea camera, or the images taken from the validated dataset, act as the training source of the newlineclassifier that trains the machine learning models to predict the sign embedded in an newlineimage of a hand pose. The accuracy of these classifier models is greatly affected by the newlineamount of noise and artifacts present in the input images, appropriate segmentation newlineapproach, adequate feature vector development and machine learning method choice. To newlineensure the high rated performance of the designed system, the research is moving in a newlinedirection to fine-tune each stage separately, considering their dependencies on subsequent newlinestages. Therefore, the most optimum solution can be obtained by considering the image newlineprocessing methodologies for improving image quality and then applying statistical newlinemethods for feature extraction and selection. The training codebook thus developed can newlinepresent the relationship between the feature values and the target class. newlineIn this thesis work, the primary goal is to design an effective model in terms of newlineaccuracy