Abstract
We propose a new feature extraction method for content-based image retrieval using a graphical rough sketch. The proposal image feature is based on the shape and the texture of the object. This edge-based feature extraction method is to represent the relative positional relationship between edge pixels. This proposed feature has advantage of shift-invariant, scale-invariant, and rotate-invariant. To verify the effectiveness of this method, we have carried out two experiments. First, in order to assess the extracted feature value, a comparison of subjective evaluation was performed using Multi-Dimensional Scaling. The result is in good agreement with human perceptual judgments. Then, as a result of applying the method to 3,200 images from COREL photo gallery, we found that the proposal feature is effective for image retrieval.
