Content-Dependent Image Search System with Automatic Weighting Mechanism for Aggregating Color, Shape, and Texture Features

Abstract

The existing image search system extracts features from the database images and performs queries thoroughly without considering the weight of each feature. Currently, all features are assigned the same weight, even though each image has different characteristics. This study proposes a new approach to image search systems that relies on content with automatic weighting. The automatic weighting process starts by calculating each moment. The first moment is obtained from the color matrix and is calculated as the average value. The second moment is obtained from the texture matrix and is calculated as the variance value. The third moment is obtained from the shape matrix and is calculated as the skewness value. These three moments are normalized to give the same weight to each feature for each picture. The results obtained for accuracy were: 70.38% for color, 60.99% for shape, 71.21% for texture, 72.65% for color-shape combinations, 78.43% for color-texture combinations, 72.65% for texture-shape combinations, and 80.5% for overall texture-color-shape features.