Efficient Image Recompress And Resize Algorithm On Very Large Scale Database

With the rapid development of the Internet and mobile devices widely used,the speed of information transmission is increasing.The reason why image information is so important is that image information is more accurate and more vivid.Because of the large number of images transmitted on the network,so it will consume a lot of network bandwidth.To avoid the network may be overload in transmitting images,we can compress the images under the condition that the compressed image looks the same with the source image.The purpose of image compressionis to transfer the orginal images as less memory as pos-sible,and preserving the image quality at the same time.This novel technology can achive faster image transmission on Internet or even in real time.Due to the sharp increasement of online pictures as well as the higher demand for image quality and loading speed,the visual lossless image compression under cloud platform is becoming the focus of the current study,so the industry has devoted itself to developing higher compression ratio and better visual effects algorithms.The paper is premised on visual lossless condition that aims to compress image as much as possible.In order to slove this problem we construct a fast compression frame which can compress massive JPEG images in real time.This paper proposes an improved JPEG evaluation algorithms with reference.Apply the evaluation function after image compression will determine wheter the image quality satisfies the requirement of visual lossless.Image as the carrier of information has to solve the second problem that how to display it on different devices which may have different resolution.As different devices may have dif-ferent width and height,image retarget is needed if we want to display an image on different devices.In this case we must as far as possible to preserve the image information.In this thesis we proposed a novel algorithm which is based on important region maintain.Firstly,we make a corner detection based on difference of Gaussian,then graham-scan algorithm is applied to do a important region detection,after that we will add different weights between important region and unimportant region.The experimental result shows our method is better than other image resizing algorithm in maintaining the important area.