Colorectal cancer is one of the common cancers in the world. Polyps are the main cause of colorectal cancer. Early detection of polyps will increase chance of treatment. In medical imaging systems, especially in digestive systems imaging, cameras carry flash lights and reflection of this light into humid surface of colon will cause specular reflection. In this thesis we proposed adaptive method based on statistical features in both RGB and HSV color space for detection of specular reflections. Our proposed inpainting method has two main stages which in the first stage we select best patch based on our proposed cost function to replace into reflection section and in the second stage we use edge smoothing method to overcome unwanted edges resulting from first stage of inpainting method. We evaluated our proposed method in detection and inapinting of specular reflection on available colon database. Our proposed method reached 99.68% of accuracy which outperforms previous works in detection of specular reflections. The method we proposed for detection of colorectal polyps uses deep convolutional neural network. We also binarized weights of proposed CNN architecture in training phase for hardware implementation in future works. We evaluated our proposed method on colon database and proposed method reached 90.28% of accuracy. Our proposed method for segmentation of colorectal polyp uses Fully Convolutional Network for segmentation of candidate polyp regions and non-linear support vector machine for post process and selecting polyp region between all candidate regions. We also evaluate our proposed method and our segmentation method reached 88.6% of sensitivity which outperforms previous method in segmentation of colorectal polyps in the same database. Keywords: Specular Reflection Detection, Inpainting of Specular reflections, Medical Image Inpainting, Deep Convolutional Neural Networks, Polyp Detection in Colonoscopy Images, Polyp Segmentation in Colonoscopy Images.
