摘要
现有的对象检测方法主要针对特定对象,当类别比较多时,难以实现实时检测与识别。提出了一种基于Objectness和梯度方向模板的大量类别下非纹理对象的实时检测与识别算法。该方法首先通过计算图像Objectness值来评价待测图像中可能出现对象的区域,大量减少可能匹配的窗口。在此基础上,在可能出现对象的区域,采用基于模板主方向和查找表的模板匹配方法,实现大量类别下非纹理对象的实时检测与识别。该方法对非纹理物体的鲁棒性较好,同时在匹配的过程中也是方向无关的。
The existing objects detection methods can not achieve real-time detection and identification when the object classes are too many. To solve the problem, a real-time detection and recognition algorithm on many classes and texture less objects was put forward. The new algorithm is based on Objectness and gradient direction template. Firstly, it eva- luates the potential objects by computing its Objectness value,which can decrease many matching windows. Then in the area where the objects may appear, it detects and recognizes the texture-less objects in many classes using the template matching method based on the main direction of template and lookup table. The robustness of this algorithm to the tex- ture-less object is better. And the algorithm is orientation independent in the process of matching.
出处
《计算机科学》
CSCD
北大核心
2015年第10期321-324,共4页
Computer Science
基金
江西省教育厅青年科学基金项目:基于稀疏编码的艺术图像语义分类技术研究(GJJ12213)
国家自然科学基金项目:大规模数据聚类的并行进化算法骨架研究(61163006)
江西师范大学青年教师成长基金:基于稀疏编码的图像语义分类技术研究(3919)资助
关键词
非纹理
模板匹配
鲁棒性
方向无关
Less-texture, Template matching, Robustness, Orientation independent