Research Of Fast Image Resizing Algorithm Based On Content-aware

With the development of the image display device,the mismatching between the images and the different display device makes the image resizing technology popular in the image processing field.Traditional image resizing methods make images distorted without considering its content.While content-aware image resizing method can diffuse the distortion to the relatively unimportant area of the image to reduce the visual error caused by the image distortion,so its scaling effect is superior to the traditional image resizing methods,but both its speed and efficiency still can't meet the real need.What's more,there is still no baseline to access the quality of resized images.According to the above situation,this paper proposes a content-aware resizing method by fusing the line-carving algorithm based on the random probability and piecewise seam carving,and puts forward an objective evaluation algorithm of image resizing.The main work is as follows:For image resizing,this paper fuses the line-carving algorithm based on the random probability and piecewise seam carving,which is a multi-operation image resizing method.After calculating the optimal threshold according to the important map and radial basis function(RBF)neural network model,images are divided into two parts,protected area and unprotected area.The two parts are resized according to image resizing ratio using the fast line-carving on random probability until the energy value of deleted or inserted seam is larger than a fixed threshold.After that,images are further resized with optimized piecewise seam carving method to meet the scale requirements.Both line-carving and seam-carving in this algorithm can effectively improve the operation efficiency and reduce the visual error caused by the twisted background area.Image resizing quality evaluation is divided into subjective evaluation and objective evaluation.For subjective evaluation,the evaluation personnel,evaluation criteria,and viewing conditions are firstly settled down,then for each image,the evaluation personnel choose two resized images with optimal visual effect from multiple resized images and save the test results.Finally the subjective evaluation is given by the statistical analysis.For Objective evaluation,according to the ground truth maps and the characteristics of the resizing algorithm,in this paper,the image content loss degree,excessively deleting edges degree and important object pieces excessively deleted degree are analyzed and fused to evaluate resized images objectively.The experimental simulation results in Matlab platform shows that the proposed fast image resizing algorithm can protect the main object in the image and reduce the visual error caused by the image deformation,most important it is faster than the state-of-art content-aware image resizing methods.The proposed evaluation method can match with the subjective perception.