Analysis of Video Content Through Object Search Using SVM Classifier

Abstract

Nowadays the content-based retrieval plays a major role in the area of research in computer vision. Although image or video retrieval is a mature technology, not much work has been done in searching of an object in video sequences. The proposed work proposes a novel method which allows a user to make queries based on visual content properties such as color percentages, layout and texture occurring in frames by using instances of prior matches. Here the author proposes a method that searches representative frames of a digital video sequence containing the required object based on input query provided by the user. The performance measures like color, texture and shape are extracted from the frames of video as well as query image to identify only those relevant frames that are matching. Color correlogram, Gabor filter and morphological operations are used to extract color, texture and shape features, respectively. The proposed work shows a good accuracy of 90% to retrieve the related frames from the video.