单元四 影像切割 II — Region Growing.pdfVIP

義守大學電機所「電腦視覺」報告 單元四 影像切割 II ─ Region Growing 參考解答 MIAT(機器智慧與自動化技術實驗室) 中 華 民 國 93 年 10 月 18 日 Region-based segmentation: 如果 R 表示一張影像的全部範圍，影像切割的目的是將 R 切割成 n個區塊： R1,R2,…,Rn 並滿足 下列條件： Seeded Region growing ： 選取一批種子 (seed)pixels ，以seed 為核心進行成長(grow) ，判斷seed周圍 pixels是否與 seed 具有 相似的特性 (灰階值 、節 理 、色 彩 ) ，如果是 ，則接受該 pixel 為同一 region ，再以此新的 pixel 為核 心，繼續偵查周圍尚未被歸類到任一 region的 pixel ，直到影像所有pixels都分類完成。 Following Adams and Bischof, we say that the seeds are grouped into n sets, A1; A2; . . .; An. Each step of the algorithm adds a single pixel to one of these sets. To achieve this and maintain a homogeneity criterion, the set T of all as-yet unallocated pixels bordering at least one region is employed where N(x) is the nearest eight neighbors of the pixel x. If for x ∈T we have that N(x) meets just one of the Ai, then the index i(x) ∈{1; 2; . . .; n} is defined such that .We define δ(x) to be a measure of how different x is from the region it adjoins. The simplest definition for δ(x) is where g(x) is the gray-scale intensity value of x. If N(x) meets two or more of the Ai, the value i(x) is taken to be the value of i such that N(x) meets Ai and δ(x) is minimized. A z ∈T is then taken such that and append z to Ai(z). This completes a single step of the algorithm and the same process is iterated until all pixels have been allocated to a set. The implementation of SRG employs a linked list storing the data of T , which is ordered according to δ(x) . Adams and Bischof refer to this as a sequentially sorted list (SSL). The SSL remains order