Research On Typical Binary Descriptors And Performance Evaluation

The descriptor of image’s partial feature is an important study question in computer visual area, a foundation of many methods, which is applied extensively in image registration, image retrieval, object identification, three-dimensional reconstruction. Recently, the mainstream descriptor of partial feature can be divided into two categories: float descriptor, bi nary descriptor. The float descriptor is floating-point type, binary descriptor is binary string represented by 0/1. Comparing with float descriptor, generated time of binary descriptor is short, and internal storage is small. But performance evaluation for binary descriptor is few. This article is mainly for study typical binary descriptor and conducting performance evaluation.The main content is as below:1. Make a study on design and implementation of the four typical binary descriptor(BRIEF descriptor, ORB descriptor, BRISK descriptor, FREAK descriptor).At the same time, conduct evaluation on ge nerated time and storage memory size, a nd make comparison with the float descriptor.2. Make performance evaluation on typical binary descriptor in image matching which includes three aspects:(1)conduct rotating, zooming in/out, lighting, noisy adding, blurring and viewpoint changing on image to get matching precision ratio, recalling rate, assumption matching rate between typical binary descriptor and float descriptor by matching original with the changed one, which aims to realize the performance evaluation on image matching of binary descriptor;(2) make evaluation on performance under the combination of binary descriptor and different detectors to get the best combination;(3) achieve performance evaluation software of typical binary descriptor in image matching.3. Make performance evaluation on typical descriptor in image searching. The article is for searching if various descriptors’ functional performance is consistent with previous evaluation result in practical application and applying typical binary descriptor in image searching based on content. In different data base, make comparison evaluation on four binary descriptor and float descriptor in four aspects, description time, search time, occupy storage size and search precision rate.