Salient region detection based on color-complexity and color-spatial features

Multivariate image data provide detailed information in variable (e.g. color, texture, orientation) and image space. Most traditional salient region detectors utilize variable information and ignore spatial information. So we propose a novel salient region detection method based on color-complexity and color-spatial features and create excellent saliency maps. The color-complexity feature is described by histogram in a quantized color space, which can effectively describe the color variation of the pixels in an image. The color-spatial feature is extracted by k-means to represent the pixel color distributions in an image. Our method is tested on a publicly available dataset and experimental results clearly demonstrate that our method integrating these two features obtains higher accuracy and efficiency than some state-of-the-art methods despite its simplicity.