A new composite histogram integrating each bin's spatial distribution for image retrieval
Hongjiang So, Yao Zhao, Baozong Yuan · 2004
In order to reduce the size of the image feature data drastically without degrading the retrieval performance, in this paper, we propose a new method to represent the image edge's statistical and distributional information. At the same time, with rich consideration of the human vision property, we improve a current color quantization scheme. Based on the above techniques, we propose a new composite histogram integrating each bin's pixel distributional information. The histogram includes only 15 bins, and the distributional information is represented by the color and edge area and position feature. Experimental results show that the proposed method efficiently reduces the feature data of images, at the same time maintaining high retrieval rate.