二值影像壓縮用可適性算術編碼

Abstract

在二值影像壓縮中，形狀在許多子主題中都起著重要作用，包括對象識別，模板匹配，圖像分析等。因此，只要我們能夠很好地編碼對象的形狀，這些應用程序的效率就會大大提高。提出的方法可以分為兩部分，無損部分和有損部分。 在有損情況下，我們使用輪廓近似技術並從學長的研究進行校正。首先，我們計算輪廓中每個點的曲率。如果該點的曲率大於閾值，則將其視為“主要點”，並且我們可以通過三階多項式近似兩個主要點之間的線段。添加優勢點的時機和位置選擇方法是我們工作中的重要問題。 但是，我們發現在某些複雜的輪廓情況下，輪廓的部分包含太多“主要點”，無法有效壓縮。在這種情況下，我們用角度弗里曼鏈碼，並利用改進的自適應算術編碼來編碼。 在無損情況下，我們不使用輪廓逼近，而是使用AF8記錄輪廓。然後，我們通過許多技巧改進了自適應算術編碼，並對AF8生成的字符進行了編碼。 最後，模擬結果表明，我們提出的方法比最新的二進制圖像壓縮方法具有更好的壓縮率。

In binary image compression, the shape plays an important role in many subtopics, including object recognition, template matching, image analysis, etc. Therefore, as long as we can well encode the shape of an object, the efficiency of these applications will be much improved. The proposed work could be divided into two parts, lossless and lossy case. In the part of lossy case, we use technique of contour approximation with correction from work of upperclassmen. First, we calculate the curvature of every points in a contour. The point will be treated as the “dominant point” if its curvature is larger than threshold and we can approximate the segment between two dominant points by a polynomial of 3rd order. The time of adding dominant point and the method of choosing position are important issues in our work. However, we found that in some of complex contour cases, parts of contour contain too many dominant point to be compressed efficiently. In this case, we apply the angle freeman chain code , and encode with improved adaptive arithmetic coding. In the part of lossless case, we do not use contour approximation, instead, we use angle Freeman chain code for 8 connectivity to record contour. Then, we improved the adaptive arithmetic coding with many skills, and encode character generated from AF8. Finally, simulation results show that our proposed method achieves better compression ratio than state-of-the-art binary image compression methods.