INVESTIGATION THE PERFORMANCE OF HARALICK'S TEXTURE FEATURES ON COLOR COOCCURENCE MATRIX FOR IMAGE RETRIEVAL

Many texture based image retrieval researches use global texture features for representing and retrieval of images from image database. Generally such researches suffer from misrepresentation of local information leading to inefficient image retrieval performance. This paper focuses on extracting local Haralick's texture feature based on predetermined region using color co-occurrence matrix method (CCM). Extensive experimental investigations were conducted to determine the best out of eleven Haralick's texture features that will provide the most efficient image retrieval performance based on precision and recall criterion. Evaluations of the retrieval performance were made based on 1000 selected images from Coral image database. From the experimental findings, it is interesting to note that for certain image categories, only six features of the eleven Haralick's texture features namely homogenity, sum of squares and sum average, sum variance, difference entropy and information measure correlation I provides the best image retrieval performance. This finding has important implication on the use of correct 'contributed features' from Haralick texture features for certain image properties as well as leading to less computational processing time due to less processing involved.