包含模糊若則法則之影像分割技術

Abstract

對於影像分割，本論文提出兩種將模糊法則加入一般影像的分割技術中。 兩種採用之分割技術為模糊 C-群群聚法，及回授式的單層感知機網路。 第一種方法，我們首先以模糊C-群群聚法分割影像。另外，根據影像之位 置及灰度值，我們提出模糊若則法則進行模糊推論，以分割影像，其分割 較重視影像朦朧區域。接著，將兩者輸出影像以‘或’算子合併，得到最 後的結果。第二種方法，以回授式單層感知機網路分割影像。結果顯示， 在朦朧區域的物體，常被誤分為背景而消失。所以，我們使用影像點的局 部標準差及垂直位置當成相關模糊集，建構模糊若則法則，決定輸出神經 元激發函數之閥值，以改進此一回授式單層感知機結構。經由測試三種紅 外線影像之分割結果，上述兩種方法配合加入之模糊若則法則，會得到一 個更好的效果。 In this thesis, two methods the fuzzy 2-means technique and modified singlelayer perceptron network, are incorporated with fuzzy if-then rules for imagesegmentation. In the first method, we first apply fuzzy c-means technique tosegment the image. On the other hand, based on vertical position and grayvalues of a pixel, a fuzzy rule inference technique is proposed to segmentthe image, with emphasis on the ambiguous regions. To obtain the final result,we combine these two output images with "or" operator. In the second method,segmented images by modified single layer perceptron network are not constantlysuccessful. We use the local standard deviation and vertical position of a pixel as fuzzy sets to construct fuzzy if-then rules. To modify the perceptronnetwork, these rules are then reasoned to determine the threshold of theactivation function of output neurons. From the results of segmenting threeimages, better segmentation images have been obtained by incorporation fuzzyif- then rules with these two segmentation techniques.