Image classification using color, texture and regions

Abstract

A new classification method using color, texture and regions is proposed in this study. Image-based features related to color and local edge patterns are used to prune irrelevant database images for each query image. The proposed region matching is then applied to find the match to the query image from among the set of candidate images in the database. The dissimilarity of each pair of images can be calculated on the basis of the matching results. Finally, all the database images in the candidate set can be sorted by ascending dissimilarity values. To achieve the classification goal, the k-NN rule is used to assign a class label to the query image. Note that the main contribution of this paper is to select proper features for representing color, texture and region, which, in turn, are used to achieve effective classification results. More important, all features used in the proposed method, no matter color or texture, are presented in the simple form of histogram, yet leading to effective results. Even in the stage of region matching, color and texture features in histograms are also used to obtain homogeneous regions and to measure dissimilarity. In addition, the proposed classification method can be applied to all kinds of color image databases rather than specific databases. The number of classes can be as versatile as required by the application. The effectiveness and practicability of the proposed method has been demonstrated by various experiments.