Image Transformation of Local Features for Rotation Invariant Pattern Matching
Mental rotation, or mental image transformation in general, is supposed to play an important role in human perception especially in invariant pattern matching and recognition. This paper presents a hypothetical process of image transformation for rotation invariant pattern matching, where we suppose human beings perform various kinds of mental image transformation for various tasks of perception, and the present process is for explaining one of them and is composed of the following procedures: first, significant local features of a memorized pattern are matched with counterparts in perceived pattern; next, with obtaining the positions of corresponding local features, the transformation is identified and then the memorized pattern is transformed and matched with the perceived pattern. Through this process, a competitive net is used for searching local features and their positions in the perceived pattern, and a method of associative memory is used for identifying the transformation. We examined the present method by means of computer simulation, and got several advantages and byproducts of this architecture.
Shuichi Kurogi Takeshi Nishida Kenji Yamamoto
Department of Control Engineering,Kyushu Institute of Technology,Kitakyushu 804-8550, Japan
国际会议
8th International Conference on Neural Information Processing(ICONIP 2001)(第八届国际神经信息处理大会)
上海
英文
731-736
2001-11-14（万方平台首次上网日期，不代表论文的发表时间）