Certain investigation on image retrieval and object recognition using intelligent techniques

Abstract

The computerized world comprises of a vast number of visual data and it is usually in the form of images and videos. The objects are considered as one of the significant information in an image. In computer vision, the recognition of an object in the image is a major task. Numerous algorithms were proposed for object recognition yet the outcomes need for betterment. The demand for image retrieval algorithms capable of finding similar images from the database increases due to the increased number of images available in public domain. In this thesis, a tailored approach for object recognition and image retrieval is proposed. Further, the Simulink model is developed for object detection and recognition. The object recognition and image retrieval systems have been developed through dissimilarity measures of object shapes, Multi Layer Preceptron Neural Network, Support Vector Machine and Convolutional Neural Network based algorithms. Various databases like MNIST hand written digits, HASYv2 mathematical symbol, COIL-20 household objects and Fashion MNIST are used to experiment the object recognition system. The image retrieval system is tested with MPEG-7 database. Initially, the object recognition is achieved through the matching of shapes between the input and reference image. The input and reference shape is considered as point set. The shape distance between the input point set and reference point set is measured using Hausdorff distance with thin plate spline transformation. Thin plate spline transformation is applied on the input point set to provide better matching with reference point set. newline