| Abstract [eng] |
Region matching between two images, displaying the same object or composition, is a difficult task. It is widely used in computer vision. Image registration, camera calibration, object recognition and tracking are only few of possible application fields. The aim of this work was to analyze and modify SURF algorithm in order to improve its recognition accuracy. SURF algorithm has difficulties in comparing differently colored images with similar features. This happens because SURF algorithm processes only monochrome images, based on image region intensities. As seen in practice, different color objects’ intensities in image can be the same, if image is converted to black and white. In that case SURF algorithm makes false image matching. In this work we studied one possible solution to this problem – image decomposition into RGB color channels before processing with SURF algorithm. “Universal video recognition system“ was created during this study and now it is used in two companies. As noted in the results of the study, modified SURF algorithm is more accurate in image comparison and responds to color changes in object’s texture. It is also better when comparing pictures in low lighting conditions. |
