In this thesis,our research goal is removal specular reflection in endoscopy using digital image processing techniques.We offer to a high-quality endoscopy image for doctor to do management of diseases in the Minimally Invasive Surgery(MIS).This thesis proposed adaptive detection method based on gray-scale value analysis and morphological dilation of specular reflection in endoscopy,as a result,the proposed method can accurately mask the specular reflection.We improved Image Inpainting algorithms.Each specular reflection area was given a category and representative parameters so that suited to the characteristics of the endoscopy image.As mentioned above we used back propagation neural network to classify the specular reflection area and specular reflection area will be inpainting.This thesis successfully applied neural network to adaptive selection parameters inpainting of specular reflection area.On experimental results,we proposed method better than original literature method and we believe that can be further attempts to inpainting more color image in the future.本發明是以數位影像處理來去除內視鏡鏡面反射,提供醫師一個更高品質的內視鏡影像,以利其對病症的處理;其最主要以適應性內視鏡鏡面反射偵測方法,根據灰階值分析,並利用形態學膨脹,準確地將鏡面反射區域標記,且改良Image Inpainting的演算法,使其適合於內視鏡影像之特性,並給予鏡面反射區域類別與各類別代表性修補參數,再以此基礎,並應用倒傳遞類神經網路進行鏡面反射區域的分類並修補,所得修補結果上皆比原Inpainting文獻方式佳。(a)、(b)、(c)、(d)‧‧‧步驟
