Seal imprint segmentation based on color feature classifier

Seals play an important role in modern society. Many researches have been investigated on the topic of automatic seal forgery identification in computer vision community. This paper puts the interest focusing on seal imprint segmentation in the real documents with cluttered background. A classification based approach has been proposed. An image pixel is classified as imprint or non-imprint depended on the output score of color feature classifier, which trained using HS-SIFT representation to the pixel under a Support Vector Machine frame. The experiments demonstrate the feasibility of the proposed method.