Combining neighbor fuzzy entropy, gradient unit feature with color-spatial feature for image retrieval

Many color-spatial based image retrieval methods have been proposed, they are efficient and effective for content-based image retrieval (CBIR). However, they loose descriptive power for images with more complex spatial layout. In this paper, in addition to color histogram, a novel edge descriptor termed neighbor quantized fuzzy entropy histogram, texture descriptor termed gradient unit histogram, and spatial descriptor termed neighbor mean histogram is proposed. These descriptors have powerful descriptive power for color image with more complex spatial layout, and are used as a hybrid visual feature index to retrieve color image. Experimental results show that this method can achieve better performance than other color-spatial based methods for the color images, especially for color natural images with relatively regular texture characteristic or structure characteristic.