A hybrid compression method based on region separation for synthetic and natural compound images

TLDR: The hybrid compression method presented can remarkably reduce mosquito distortion around the contours of character or graphics and provides good performance for the compound images compared with JPEG.

Abstract: This paper presents a hybrid compression method for a compound image including characters, graphics, and photos. Such kind of image is popular as Web contents or TV game images. We use usually JPEG to compress it, but the DCT compression like JPEG causes strong mosquito distortion around the contours of characters or graphics because of their strong edge intensity. Therefore we need high quality and low bitrate coding scheme for the compound image. Our hybrid compression method is based on region separation and adaptive coding. This method separates synthetic image region and natural image region according to each features, i.e., edge intensity and color deviation. Then our method encodes synthetic region by runlength coding, and natural region by JPEG respectively from the viewpoint of visual quality and coding efficiency. Our hybrid compression method provides good performance for the compound images compared with JPEG. This method can remarkably reduce mosquito distortion around the contours of character or graphics. Experimental results show +5 to +10 dB improvement of the SNR over JPEG.