Research And Improvement Scaling Algorithm Based On Seam Carving Image

Image resizing has been a focus of research in the image processing field. Geometric transformation resizing method uses uniform model to resize the entire image, considering of no contents of the image, so it can not maintain the edge information and details of image well. Seam Carving algorithm is an algorithm based on image content, compared with the geometric transformation resizing method, it is better to maintain the image details of the feature region. But it still causes deformation in the important parts of images in some degree. This thesis works on based on Seam Carving algorithm.With researching the image processing technology more and more deeply, more research achievement produced in image saliency. Image saliency defines the area of human visual attention of the image by simulating human visual system. In order to improve the deficiencies of Seam Carving resizing algorithm, this thesis proposes a method of using the image gradient map and saliency map to define the energy function, namely using the edge detection algorithms and GBVS model to calculate the image gradient map and saliency map, and then normalizing the image gradient map, and then comparing the value of normalized saliency graph and the gradient map corresponding to each pixel, selecting the larger one as the new value of the energy map, generating a new image energy map, then using the dynamic programming algorithm to find the optimal pixel line, determining the sequence of the optimal pixel line.In order to verify the advantages and maneuverability of proposed idea, selecting three pictures as experimental images, respectively using Seam Carving algorithm, Optimized Scale and Stretch method, mesh deformation method based on saliency and the method proposed in this thesis resizing the same columns, and then comparing and analyzing the results. Evaluate the effect on visual quality and the average energy value of resized images. Experimental results show that, in these two aspects, the improved method gets a better effect than the Seam Carving algorithm as well as the other two algorithms.The results of method proposed in this thesis show that define the energy function of image by gradient map and saliency map can not only effectively maintain the image edge information, but also to keep smaller distortion in the region of interest of human visual. Gradient map and saliency map combines well in the Seam Carving algorithm, it has obviously practice value.