Research On Image Resizing And Similarity Criterion

With the rapid development of the multimedia technology, a variety of display devices emerge endlessly, which requires digital image should be changed to different sizes for displaying, and the integrality and visuality of the important and interesting information are protected concurrently.Seam carving realizes the image resizing through removing or duplicating the unimportant regions. So it can protect the important regions of image beautifully, and obtains good resizing results. The detail process of Seam Carving is studied. The optimal 8-connected seams of pixels are searched by Dynamic Programming firstly, and then remove or duplicate these seams to realize image resizing.Bidirectional similarity measure is studied, which computes the dissimilarity between images from coherence and completeness two sides. Based on this algorithm, the dissimilarity values and its relative difference between the resized image and the original image are computed going with the image resizing. Experiments show that there are some relations between the relative differences of dissimilarity and the degree of image distortion. On the basis of above the similarity criterion for image resizing is concluded in this paper.Based on the similarity criterion, a new image resizing algorithm is studied in this paper. It combines Seam Carving with Bidirectional Similarity measure, to obtain better resizing results. First implement image resizing by Seam Carving step by step, and for each step remove certain seams, meanwhile calculate the dissimilarity between the resized image and the original image, as well as the relative difference of dissimilarity between neighboring steps. Then according to the relative differences of dissimilarity in each step, based on the similarity criterion, the degree of distortion of the resized image can be assessed. Transform the resizing method from Seam Carving, before the salient feature being damaged markedly, to Bidirectional Similarity Iteration Optimization, and resize image up to a desired size. Experiments show that the resizing algorithm researched in this paper can preserve both the important regions and the global visual effect of the original image, and also can avoid the disorder and distortion of image contents.A system of image resizing has been exploited based on the above-mentioned research. It can realize image resizing on the basis of serval different algorithms and the computation of dissimilarity bwtween images.