Fuzzy classifier with support vector learning for image retrieval using a specified object

This paper proposes a specific object based image retrieval method using a Batch Support Vector Machine-Trained Fuzzy Classifier (BSVM-FC). The goal of this method is to retrieve images containing the specified object from an image database. This can be considered as an object detection problem. The interested object is assumed to contain multi-colors in non-homogeneous distributions. The feature used for detection is color histogram of the object appearance composing pixels on the Hue and Saturation (HS) color space. To obtain distinctive color histogram, the HS space is non-uniformly partitioned. Based on the histogram, the BSVM-FC is used as a detector in order to improve detection performance. The consequent part of the BSVM-FC is learned through a linear support vector machine in order to give the fuzzy classifier better generalization performance. Experimental results on image retrieval using a specified beverage can and comparisons with different classifiers show the advantage of using the BSVM-FC in this retrieval application problem.