Notes Image Classification

Bill Image Classification is a very important process of an Automatic Bill Image Processing. It determines the bill image's type by the analysis of the bill image's layout part feature, and provides a priori layout information for the latter automatic recognition. In this paper, we research and improve some common technologies of Bill Image Classification to design a complete Bill Image Classification System through extracting bill images'frame lines feature and titles feature.The Bill Image Classification System designed in the paper is mainly composed of three parts such as Line Matching Classification, Title Matching Classification and Title Recognition Classification. First we have the image which is to be classified and frame line template library for Line Matching Classification; Then we use the method of Title Matching Classification to sort bill images which are uncertain of the type or refused to be recogonized according to Line Matching Classification; Finally we use the method of Title Recognition Classification to sort the bill image.Line Matching Classification is a rough classification process for the bill image, including three stages such as the preprocessing, the frame line detecting and the line matching. Among them, preprocessing is composed of print removing and lines binarizing; The line detecting uses chain code as the basis, combined with the frame lines'position relation to detect a certain length of the horizontal and vertical lines in the bill image; Line Matching s the matching degree of bill frame lines and the template frame line through using the defined similarity matching model, then we can the determine bill's type according to matching degree.The title matching and title recognition sort the bill image according to the title feature. First we locate the area of the bill image's title; Then we calculate the matching degree of title image and template image by using the image matching algorithm, and determine the type of the bill according to the matching degree. In order to increase the speed of the matching algorithm, we improve the matching degree calculation and matching strategy, and the experimental results show that this algorithm is effective.Title Recognition Classification is the Title Matching Classification's supplement. First we locate the area of the title accurately; Then we use the directional projection to divide characters; Finally we extract the character's contour feature and average line density feature for title recogonizing, according to which we determine the bill's type. The experimental results show that, the method of using a single method to sort bill image have yet to be improved. But after combining the three classification methods, we obtain good experimental results on a small sample set.