IR-FF-kNN: Image Retrieval Using Feature Fusion with k-Nearest Neighbour Classifier

Abstract

Presence of inconsistency in the visual appearance of image tends to degrade the retrieval process performance. Increasing image data across several domains encourages to explore visual information of image representation to simplify the interpretation and concentrate on discriminative features of images so as to use them for retrieving relevant images for increasing machine learning model performance. Thus, features fusion and k-Nearest Neighbours (IR-FF-kNN); an Image Retrieval framework is proposed to increase retrieval performance. The Histogram of oriented Gradients (HoG), Color Moments (CM) and Center Symmetric Local Binary Pattern (CSLBP) descriptors are used to obtain multiple features of images in the features extraction phase and in similarity computation phase, the kNN classifier is used. The proposed framework is tested on MIR Flickr dataset and provides mean average precision of 85% compared to the state-of-the-arts.