Research On Digital Image Resizing And Quality Assessment

Getting the best display effect of full screen in display applications,the input image which resolution is difference should be zoomed to match the physical resolution of the terminal display.And the image resizing algorithm directly affects the quality of the displayed image.In digital image processing system,testing the performance of algorithm and system is inextricably linked with the image quality assessment.Therefore,digital image resizing and image quality assessment as the key technologies of digital image processing are becoming more and more attention.Under the actual program,this paper researches the image resizing technology and quality assessment,including:1.A matching re-sampling filter algorithm was proposed for the problem of re-sampling scaling techniques in front-end equipment of acquisition and compression.The algorithm that based on the error minimization principle and obtained parameters of the down-sampling filter derives the corresponding up-sampling filter to achieve the matching purpose.It can improve not only the image quality after decoding,but also the encoding performance.These are conductive to the post-processing.Compared to the simulation results of JSVM,the proposed matching method has obtained the improvement performance and the PSNR is significantly raised.2.Two novel scaling mechanisms were proposed for the problems in classic and some adaptive image interpolation methods.These algorithms often suffer from visual artifacts,complex model or slow interpolation speed.These two novel methods are edge-directed smoothness algorithm and area-directed image interpolation based on threshold.The former proposed method based on the combination of the edge-directed smoothness filter and the bilinear interpolation algorithm not only can enhance the enlarged image edges effectively,but also make the interpolated image get higher quality both in edges and smoothness areas.The latter proposed algorithm exploits the threshold-judgment to classify the regions around the interpolated pixel.The region that the interpolated pixel should belong to was decided by threshold combined with the nearest neighbor and the mode method.According to the different types of interpolated pixel,the linear interpolation formulas with different weights were designed to achieve adaptive interpolation and improve image quality.The simulation results show that,edge-directed smoothness algorithm compared to the edge-adaptive interpolation method,the interpolation speed improved significantly,but compared with classic algorithm,it can effectively reduce the jagged edge and fuzzy edge phenomenon,and get improved image quality with better visual effect.The area-directed image interpolation method solved the image distortion problem caused by processing edges poorly in classic methods.Compared with the edge adaptive image interpolation algorithm,the area-directed image method improves the interpolation speed significantly under the premise of without significantly reduced image quality and obtains the interpolated image with high objective quality and visual effect.3.Two novel image non-proportional scaling mechanisms based on content-aware were proposed for the problems in classic content-aware image resizing methods.These algorithms often suffer from high calculation complexity,slow operational speed and have no universality.These two novel methods are content-aware image resizing based on random permutation and optimize resizing method with merging and improved importance diffusion.The former proposed method exploits visual salience to get a random scrambled energy vector,which is extracted uniform to protect image important content.It can efficiently avoid the local distortion caused by excessive deleting non-important area.The latter proposed algorithm realizes image reduction and expansion by merging the two most unimportant rows/columns and inserting a new one between the two most unimportant rows/columns to preserve important contents.To avoid the local distortion caused by excessive merging non-important area,the energy of the merged one was proposed and diffused to four rows/columns around it.The simulation results show that,Compared with the seam caving algorithm,the resizing method based on random permutation increases the operational speed effectively.The resizing scale is larger,the average amplitude of operational time reducing is more obvious.But there also exists image distortion in non-important region.Compared with the seam caving algorithm,the resizing method with merging and improved importance diffusion increases the operational speed and avoids local distortion.It also can obtain image with high objective quality and visual effect,and it is suitable for various types of images.4.A novel mechanism based on space domain natural scene statistic was proposed for the problems in the state-of-the-art no reference image quality assessment methods.These algorithms often suffer from the poor universality,complex model or high calculation complexity.The logarithmic statistical features were introduced into the proposed algorithm.To make an evaluation to the testing image,the proposed algorithm makes use of measurable deviations from statistical regularities observed in natural images.The simulation results show that the proposed algorithm gets a good consistency with human subjective perception.Compared with the current no-reference methods,the proposed algorithm is simple and effective,and is suitable for various distortion types.