In this dissertation, the problem of image processing is studied in the view of visual control. The image processing algorithms are discussed for the pavement distress images and welding seam images with heavy noise. The key technologies are simply analyzed for the autonomous working of robots. 1. A new approach is presented for the problem of pavement dilapidation classification based on Markov Model. In this approach, a picture is partitioned to blocks, and every block has a state with local shape information. Then the state probability and the state transfer probability are computed for classification. The effectiveness of the algorithm is verified by experiments. 2. A new edge detection approach based on Radon transform in sub-areas is designed for welding seam images with noise of severe disturbances. It is divided to many sub-areas for a frame of welding seam image. Edge detector is applied to sub-areas, and candidate welding seam edges in a sub-area is recognized via Radon transform according to the knowledge of the welding seam and the disturbances in advance. Based on the candidate welding seam edges detected from sub-areas, Hough transform is employed to extract the feature lines of the welding seam. The comparison experiments with traditional methods are also provided to illustrate the performances of the proposed method. The experimental results verify the effectiveness of the proposed approach. 3. The configuration and classification of visual servoing systems are introduced. The key technologies such as the selection of image features and the design of control algorithms are simply analyzed.