Research On Compression Algorithm Of Image Local Features

Image compression is an important part of digital image application which has always been a hot topic. Currently most image coding technologies compress images by pixel. However, this method may not be able to use the external images, especially the highly correlated resources in the cloud. Faced with such problem, some scholars proposed a new compression method——Cloud-Based Image Coding in2013. This method applies image feature description to find highly correlated images and then reconstruct them from the resources library in the cloud. In this paper, based on the Cloud-Based Image Coding framework, we improved the compression algorithm of residual feature vectors by two steps. Firstly, reduce the number of residual feature vectors which remain to be encoded. Secondly, improve the residual feature vectors encoding algorithm. The experimental results showed that the improved algorithm performed more excellently on both the compression efficiency and visual effects.The main work is as follows:study the Cloud-Based Image Coding framework, including the extraction of local features, the compression algorithm of local features and other key technologies of the framework. The core of the research is the analysis of the effect of compression rate and visual quality caused by different transformation, quantization and compression algorithms. In this paper, our improvement consists of three parts, which are listed as follows:(1) Screen for features before encoded to reduce the amount of data through reducing the total number of features.(2) Screen for residual feature vector to reduce the number of residual feature vectors to be encoded, to further reduce the amount of data. (3) Improve the coding algorithm of the residual feature vector to reduce the amount of the residual feature vectors, in order to improve the feature matching degree and the subjective visual quality of the reconstructed image.The experimental results show that the images’average compression ratio of our optimized and improved algorithm increases about eight percent, and the subjective visual quality of the image is also improved.