Research On Binary Indexing Technology In Large Scale Image Filtering

It is necessary to filter the sensitive images from internet in order to create a good network environment. Therefore, image filtering is an important research direction in the field of information content security. The general process of image filtering is:firstly, extracting features of images to be detected and describing these features in high-dimensional vectors; secondly, querying in the library which is consist of high dimensional vectors of sensitive images; lastly, filtering sensitive images according to similar content.An efficient indexing technology is the key to achieve large scale image filtering because of high-dimensional image features and image filtering which is faced with large scale images. After decades of research, the current indexing technology is facing the following problems:query performance declines because of high dimensional data and large-scale data leads to insufficient memory resource. Existing research can’t be able to effectively solve the above problems, therefore how to organize the large-scale data and make similarity query rapidly and accurately is a very challenging research focus and difficult problem.Recently, binary indexing technology is an active topic, because the binary features occupy less memory and can be matched rapidly. In order to achieve real-time filtering of large-scale images, we propose a binary hierarchical indexing technology. The main contributions are summarized as follows:(1) Propose a method to compress binary features based on standard deviation. We measure the distinction of each dimension according to the standard deviation and generate binary features which are short and containing important information. This method decreases the consumption of memory resource.(2) Conduct experiments and improve a clustering method which is the fusion of division and hierarchical decomposing methods. This method clusters the binary features efficiently and lays the foundation for building a binary index.(3) Propose a binary hierarchical index algorithm. Firstly, we choose bits of high distinction from binary features to compose clustering dataset; secondly, we cluster the dataset and build binary hierarchical index tree which decomposes the search space hierarchically; lastly, we query binary features by adopting preliminary filtering and further filtering which not only ensures the filtering accuracy but also improve the filtering speed.Based on the above research results, we build the image filtering system and make experimental analysis of binary hierarchical indexing technology. The experiments show that our method significantly improves the query speed, ensures filtering, accuracy reduces the consumption of memory and meet the needs of the mass real-time image filtering.