Feature Extraction Based On Edge Detection And Corner Detection And Its Applications

Image feature extraction plays an important role in pattern recognition, artificial intelligence and computer vision. We can define different features in different applications. Generally speaking, color, texture, edge and corner are the main features. A plenty of methods for extracting features, including edge detection, corner detection and texture analysis, have been proposed in the last two decade with the advances of digital image processing. Some detectors become popular thanks to their good performance.An edge detector and a corner detector were presented in this thesis based on the existing detectors and the theory of wavelet and multiscale geometry analysis. The ridgelet transform frame based on continuous ridgelet transform was discreted to adapt to digital image processing. At first, applying discrete Radon transform to image to obtain the Radon domain, and then applying discrete wavelet transform to the Radon domain, we get the discrete ridgelet transform of the image. Applying discrete ridgelet transform to image mainly containing liner edges, then the edges can be localized using the corresponding linear parameters of the sparse coefficients in ridgelet domain. We proposed a corner detector based on local edge points: For each edge point in a local square window, a "corner measure" was defined using the "arc-chord" distance, and then the points of local maximum can be found using nonmaximum suppression. We call these points of local maximum "corner measure" as candidate corner points. Fitting both side edge points of the candidate corner point with two lines, we can calculate the angle between this two lines. By comparing this angle to a given threshold, we can determine whether the candidate corner point is a corner point or not. The two detectors show good performance when applying to some images rich in features respectively.The paper can mainly be divided into two parts: in the first part, a review of feature extraction was introduced; various detectors for extracting different features were explained in detail, especially to some classic detectors; two detectors proposed by author were also presented; as the field closely relating to feature extraction, we also briefly discussed digital image processing and image understanding.Some of the applications of feature extraction were given in the second part. Feature was widely used in many applications such as pattern recognition and scene analysis, it is a fundamental constituent in visual image processing. But few problems make it as the final result, in most cases it was used as an evidence for further analysis, such as image segmentation based on detected edge points and image registration based on detected corners and so on. In this part, several simple applications of feature were given; the perspective of feature extraction was also discussed based on the author's understanding.