Comparing Bag of features Techniques for Image Retrieval

Abstract

Bag of Feature (BOF) is a customizable technique that provides a free hand on the type of feature selection for creating visual-word vocabulary that fits better for the various image retrieval applications and this is the main advantage of using this technique to create a customizable feature mix of local and global features. One of the two prominent techniques for creating Bag of features in image processing and computer vision is through grid formation and custom extractor. Therefore, selection of type of feature used for Bag creation is very basic and critical one as whole process of Retrieval depends upon it. Feature Color is the well known one to be used most. In this paper L*a*b color space is used for creating one of the feature Bag as L*a*b Colorbag. While two other KAZEbag and SURFbag are created using different descriptor methods namely SURF and Kaze. For unfolding different hidden patters out of collected visual words some machine learning intelligence is adopted like support Vector Machine is used here for clustering and creating cluster centers. MATLAB R2021a is used for all the implementation. Along with the Dataset of Natural images with ten categories is used for evaluation. All the three created Bags are compared on the different parameters such as number of features extracted, strongest feature metric, word frequency and average precision.