Design And Implementation Of Image Stitching Based On Corner Detection And Matching

Image stitching is a key technology in expanding the application of image view. However, nowadays, the object of the research in image stitching is below800×600. In order to improve the practicability, an image stitching method based on corner arming at dealing with large resolution image is implemented in this paper.The study of image stitching is based on frequency, image blocks, feature field, and image feature combined with network both at home and abroad. But these technologies are more complex than the method that grounded on feature points. So this paper implements image stitching with higher resolution based on improved Harris corner detecting and matching. The main job in this paper is showed as below.In the preprocessing of image, firstly use image to gray image is gained by function transforming gray. In the Harris corner detected and extracted, based on the traditional algorithm of Harris corner detected, the algorithm is improved in the term of differentiate operator, Harris reflect function and corner threshold R. In image stitching, firstly, executing coarse matching, and then refining the corner using accurate matching. Coarse matching is achieved by using the sum of square of pixel difference and normalization cross correlation and the corners is refined by using random sample consensus based on gradient of disparity constraint. Coarse matching can resolve the Harris corner problem brought by light in a better way. Random sample consensus based on gradient of disparity constraint can redetect the sample corners before seeking the answer of model. This can reduce the false matched corners. In this way, the time brought by false data can be saved without decreasing the precision. In image merging, average image can be used to balance the light difference between two images, and there is no obvious change in the same fields between two images after merging. The image merging without gaps is implemented by using smooth weights.The corner detecting and image stitching system is accomplished under visual studio2008develop environment. The algorithm becomes more sensitive to gradation change, and more accurate to corner position, which avoids the influences of the coefficient k in response function. The coarse stitching can solve the problem which produces more Harris corners. This system implements the patching of two colorful images with size1024×1024, and the result meets the requirement of image patching on the whole.