Transformational approach for alignment-free image matching applications

One of the fundamental steps in computer vision and image processing applications is the alignment of images before matching. A number of alignment methods based on features points, pixel information and transformation parameters have been proposed in the literature. All these methods either suffer from accurate detection of feature points or are computationally expensive that makes the entire matching process time consuming. The fast matching of images performed without alignment can benefit a number of image matching applications. In this paper, a transformational approach is proposed that identifies whether two images are similar or not without performing an alignment. The matching without alignment reduces the computation time of the entire matching process. The effectiveness of the proposed approach has been demonstrated on UMD logo dataset. The robustness of the proposed approach to additive noise shows that high accuracy can be achieved even when the noise is as high as 20%. Furthermore, the computation time is improved by decreasing the number of projections without affecting the matching accuracy. The experiments demonstrate that the proposed method is not only efficient but also highly accurate as compared to existing approaches.