Registration Of Color Images Based On Color Invariants And Affine Invariant

In recent years, many scholars research on image registration and put forward a series of algorithms. ASIFT algorithm has been proved to be a fully affine invariant image registration algorithm. Compared to the classic scale SIFT algorithm, ASIFT algorithm adds two parameters are based on SIFT algorithm which related to the camera angle, simulate as many different camera angles set of images for registration images, then using SIFT algorithm to detect registration for the simulated image sets with exhaustive attack method. Therefore, ASIFT algorithm not only has the advantages of the SIFT algorithm, that is, the invariance for rotation, scaling and translation transformations, but also has the invariance for shear transformation in Affine transformation, thus achieves fully affine invariance. Even if there is larger perspective transform between images, the ASIFT algorithm can still achieve reliable registration.The color information has played a very important role in the description and identification of objects. In the registration process for color images, if algorithm considers only the grayscale information and ignore the color information, likely to lead to the wrong match. When ASIFT algorithm register for color images, first, we converts the color image to grayscale, then we extracts feature points according to the image gray level information, thus, resulting in a lack of color information. When lighting conditions change between registration images, the prospective effect of ASIFT algorithm dropped significantly.Color invariant theory provides a method for color image registration. The extracted color invariants for color images based on color invariant model, has nothing to do with the observation position, surface orientation, illumination direction, intensity, and Fresnel reflection coefficient. Therefore, we combined the color invariant theory and ASIFT algorithm, proposed a new color image registration algorithm CASIFT. Based on the ASIFT algorithm, we use color invariants to extract feature points instead of the gray-scale information and establish the characteristics according to the color information of the surrounding area of the feature points, the problem for ASIFT algorithm can not match images under different lighting conditions is solved. Experiments show that, combining the color variables, ASIFT of algorithm is more robust in color image registration.