Cigarette pack image segmentation based on differential projection

Image based recognition has been applied in different fields. Lots of machine learning and artificial intelligent technologies have also been used for data driven study. However, recognition with less data still remains challenge. In terms of the cigarette image, this paper proposes an automatic segmentation and recognition algorithm. First of all, a background gray value automatic estimation method is proposed, with which the image contrast is then be enhanced. Subsequently, by using the estimated background gray value, the region detection is realized on the image projection signal. Thirdly, differential projection is applied to detect the logo area. Finally, the color, geometrical and texture features are extracted for recognizing the type of cigarettes. Experiments verify that the proposed method is effective.