Image matching is a basis in the image processing and computer vision field, it comes from a number of practical problems. The primary task of Point feature matching is to extract the stability features, and to describe them. These features can maintain certain invariance to the image changes factors, such as rotation, zoom scale, affine transformations, and perspective changes, light changes etc., to the object movement, occlusion, noise and other factors can also maintain good matching. A point-based scale and rotation invariant feature transform algorithm for the sub-pixel image registration is proposed. First of all the feature detection and identification are made in the scale space, the position and scale of key points are determined, and then use main direction of the neighborhood's gradient as the direction of the key point in order to achieve the independence of the operator of the scale and direction. The method is invariance to rotation, scale scaling and brightness changes, else maintaining a certain degree of stability to changes of perspective, affine transformation and noise. It is unique, informative and suitable for fast and accurate matching in the mass characteristics database. Being high-speed, the matching algorithm can achieve real-time requirements.