ABSTRACT
It is difficult to deny the existence and domination of images in every domain, especially their use over the internet on almost all applications. This bombardment of image data into the online repositories has caused the rise of various much needed techniques to allow the user to handle image-related operations with ease, speed, and accuracy. Therefore, the possibilities of research in this area have opened up. However, many domains, including image classification and retrieval, have become in high demand. In the past few decades, constructive work has been done, especially in automatic feature detection and extraction using Machine learning techniques such as Bag of Visual works or Bag of feature, which are prominent and referenced here. The images are subjected to a system where automatic features are extracted and quantized so that a visual vocabulary is created. This vocabulary drives the process of retrieval of the most promising matches with respect to query images. In this work, along with the BOVW techniques, different feature detection and extraction methods are deployed, like SURF, FAST, BRISK, MSER, and ORB, along with the end results. Also, FAST results are compared using different similarity measures. All this methodology is implemented using Matlab TM on the WANG dataset of 1000 images, with 10 different categories of 100 jpg images each. The proposed technique performance has turned out to be 90% approximate for each category.
