含雜訊之線狀影像的骨幹擷取

Abstract

線狀二值影像的骨幹能反映物體的顯著幾何形狀，對於物體構形的描述、 辨認與儲存，具有很大價值。然而，影像難免會被雜訊干擾，其所造成的 問題，主要有破洞、邊界扭曲、斷裂、碰撞等。這些問題不易克服，致使 所得骨幹巳嚴重變形，無法再顯現物體原本的形狀。本文將分析前述各項 問題，並採用硬體方式做前置處理的大量計算，以節省執行時間；對於前 置處理所得的結果，再做高階的軟體處理。我們將這些處理實際應用到多 種影像資料，以測試處理效果是否符合預期目標，依所獲得的成果研判， 絕大多數都能得到不錯的結果。 The prominent structural shape of a line-shaped binary object image is generally kept in the skeleton of the image. This skeleton can be very valuable to the object description, recognition and memory storage, etc. However, the inevitable noise arising during the imaging and processing processes creates small holes, rough edges, gaps, and accidental touching, etc. These factors lead to the serious distortion of the skeleton that no longer preserves the true shape of the object. It becomes very desirable to solve these problems. In this thesis, the foregoing problems caused by image noise will be handled in two ways. First of all, preprocessing operations such as removing accidental touching or holes and smoothing the boundary will be implemented in hardware. The thinning operation is also implemented in hardware. As for the high-level operations of connecting two broken lines, breaking-up lines meeting at the fork points or bridge points, they will be solved by software techniques. We test our hardware and softaware methods with real images. The computer simulations show satisfactory results that almost live up to our requirements and expectations set up at outset.