Image Feature Point Matching Algorithm Research Based On Image Enhancement

Image feature matching is a hot research issue in the related researchfields, such as image processing, computer vision and3D reconstruction. Imagefeature is the key to determining the similarity and the effect of classification. The keyissue of the cognition and the identification is to find the suitable image feature.Image feature matching has the strong ability in view of the viewpoint change, theaffine transformation and noise etc, and it has lower computation cost. Therefore, it iswidely applied in many fields. At present, some image feature matching algorithmsthat include SIFT, SURF and ASIFT have a good stability with scale, rotation andillumination change, and had been improved in the study. In the process of the imageacquisition, the different imaging conditions (the position, posture of the camera andlight etc) and the performance of the sensor itself make different appearance of thesame area in different images, and the correspondence area of the images may havethe projective distortion. However, the matching of image feature points is thefoundation of the subsequent image processing. A few points matched and lowmatching accuracy can influence the subsequent image processing, such as, imageregistration, image mosaic etc. The main research work of this paper is to increase thenumber of image feature matching points. The imaging condition could result in thepoor visual effect. Therefore, the poor images need to be preprocessed to enhance theuseful information of the image, and increase the number of image feature matchingpoints.In order to achieve this purpose, histogram equalization, homomorphic filter based onlight and reflection model, and the image enhancement method based on the Retinextheory are applied respectively to preprocess the blur image and the image withillumination change. The experimental results show that the image is enhanced, andthe visual effect is improved obviously. Then, these images are processed respectivelyby three classic matching algorithms of the image feature points (SIFT, SURF andASIFT). The experimental results corresponding to the nine combination methodsshow that the number of the preprocessed image feature points is increased obviouslyby comparing with the original images. The image preprocessed by the histogramequalization can more enhance effectively the number of the image feature matchingpoints. In addition, for the blurred image, the matching rate of the image featurepoints extracted by the SURF algorithm is higher in the histogram equalizationpreprocessing. While in the homomorphic filter preprocessing, the matching rate ofthe image feature points extracted by the SIFT algorithm is higher. What is more, theimage preprocessed by the histogram equalization can more improve effectively thematching rate of the image feature points than the other two enhancement algorithmsfor the image with illumination change. At the same time, the good performance of the original feature matching algorithms can be inherited, such as, the rotationinvariant, the scale invariant and the affine invariant and so on.