Extracting Image Dominant Color Features Based on Region Growing

Abstract

Dominant color is a compact and efficient descriptor which employs representative colors to characterize the color information in the interesting region of an image. Dominant color descriptor is suitable for representing local features of images and can be used for quick retrieval in large image databases. A method for extracting image dominant color based on region growing algorithm was proposed in this paper. The steps of proposed method are as follows: Firstly, the color space of an image is transformed from RGB to HSV; Secondly, region growing algorithm is used to divide an image into color regions, and regions of which the areas are less than 1 per cent of the image will be filtered out; Thirdly, neighbor regions with similar color are then merged, and if area of the merged region compared with the image is more than 5%, its color will be considered as a dominant color. Proposed method was compared with well known color histogram based algorithm in this paper. Two color evaluation criterion proposed by MPEG-7 standards, i.e. ANMRR (average normalized modified retrieval rank) and ARR (average retrieval rate) were used in experiments respectively. According to MPEG-7 standards, less ANMRR or larger ARR indicate a good image retrieval result. For a 468 images database with 4 differences similar image groups, ANMRR and ARR of proposed method reduced 10.7% and increased 7.46% respective compared with color histogram based algorithm. The experiment results showed, according to ANMRR and ARR, the performance of proposed method is better than that of compared algorithm.