【摘要】 局部图像特征描述是计算机视觉中一个基本研究问题.二值描述子采用位存储并以汉明距离作为其匹配准则,在内存占用率和匹配时间上相比传统实值描述子有很大的优势.色彩属性是图像的一个重要属性,对于旋转、尺度变换、模糊以及轻微的光照变化都具有不变性.提出一种基于局部灰度分布和色彩属性的二值描述子.提出的描述子融合了局部灰度分布信息和色彩属性,使用灰度信息进行初步匹配后,利用特征点的色彩属性筛除误匹配.在Oxford sets数据集上测试,该融合描述子可以获得与USB(ultrashort binary descriptors)同等的精确率以及比USB高20%的召回率.实验表明,与SURF(speeded up robust features)等特征相比,新描述子的速度快、内存小,更易满足实时场景和移动设备的需求.更多还原

【Abstract】 The description of local image features is a basic research problem in computer vision. The binary descriptor employs bit storage and uses Hamming distance as its matching criterion,therefore it has a great advantage comparing with traditional float descriptor in terms of memory usage rate and matching time. Color attribute is an important attribute of the image,which has invariance to rotation,scale transform,fuzzy and slight illumination change. This paper proposes a binary descriptor based on local gray distribution and color properties. This descriptor fuses local gray information and color properties. The gray binary feature obtains local gray-level information from the neighborhood of key point and realizes rough matching by using gray information. As complementary color information,the color binary feature filters false matches later. Testing on Oxford sets,fusion descriptors can obtain precision rate nearly equal to USB( ultrashort binary descriptors) and 20% higher recall rate than rawUSB. Contrasting with SURF( speeded up robust features),the fusion features possess faster speed and smaller memory cost. Therefore it is more likely to meet the needs of real-time scene and mobile devices.更多还原