Content-aware Image Resizing And Image Rotation

Recently, display devices with different aspect ratio have been widely used due to the rapid growing of intelligent terminal equipment. We should resize the image when adapting an image to different display screens. But the traditional scaling method usually distorts the objects in the image. Therefore, image resizing attracts many researchers, and is becoming a research hotspot.Currently, the image resizing methods are mainly seam carving and image warping. This paper presents an improved mesh parameterization approach based on the significant map. The method can be implemented in two steps:Firstly, compute the adaptive significant map by combining the salient map, gradient map and scale together, which describes the content of the image. Then, partition the image into grids by horizontal and vertical lines, and establish the grid distortion function based on the adaptive significant map, which is globally solved by an iterative method. In each iterative step, we simply solve a set of linear equations. Besides, we extend this method to irregular image resizing, and give important region preserving approach by modifying the boundary constraints. Experimental results show that our methods can efficiently uniformly scale important regions by reasonably distorting unimportant regions in resizing an image.As we know, image rotation will lead to jagged edge and image blurry, especially for dominant structure images. We propose a content-aware image rotation method, which efficiently improves the rotation results. Firstly, use the Sobel operation to obtain the gradient map which depicts the object structure of the images. Second, take gradient map as weight, and use weighting bilinear interpolation to rotate image. This method combines the gradient value with the distance to make weights when processing the pixels in the sharpness regions. Compared to conventional method, the time complexity of our method is much lower than bi-cubic method and similar to bilinear. Owing to taking the structures into consideration, our method can achieve better performance in rotating images.