| The size of the image set grows explosively,and the storage cost is becoming more and more expensive.In order to save storage space,it becomes necessary to compress storage space.Fortunately,there will be a lot of redundancy between similar images.Find these similar images and remove the redundancy,then we can reduce the amount of storage space.Based on the research on the method of compression based on image features and the analysis of image retrieval field,an image set compression method based on inverted index is proposed.It uses the inverted index which is often used in image retrieval to get similar image sets.Then it uses the video stream to compress;The image set compression method based on inverted index is as follows: we first use kmeans clustering algorithm to generate the codebook,and then get the index.In order to improve the accuracy of the index,an improvement of the traditional kmeans algorithm is proposed.A bit vector is added to each image feature.By measuring the Euclidean distance and the Hamming distance between features,we sort the pictures.It ensures the higher accuracy of the retrieval results;At the same time,the knn algorithm and the bit vector are used to improve the processing speed of this method.We need to make the pictures ordered which are in similar image sets that are generated based on inverted index.Then according to the ordered pictures,we can use video stream to compress the set.This paper analyzes the current research situation of image rearrangement,and makes three improvements to the image rearrangement process: cross matching and ransac algorithm are used to improve the authentication precision of feature point matching pairs.Using degree and weights to get a reasonable head node is adopted.Using threshold to determine the pruning program is also proposed.According to the test result,it can be done in the speed and compression ratio on the balance and the value of psnr can also reach the expected. |
PDF Full Text Request