A human-perception-like image recognition system based on PAP vector representation with multi resolution concept
A human-perception-like image recognition system based on PAP vector representation with multi resolution concept
复制标题
基于具有多分辨率概念的 PAP 矢量表示的类人感知图像识别系统
DOI：
10.1109/icassp.2002.5743974
复制
发表时间：
2002
期刊：
影响因子：
--
通讯作者：
T. Shibata
中科院分区：
文献类型：
--
作者：
M. Yagi;T. Shibata
一个类似人类的强大的图像识别系统的灵感来自心理的大脑模型[1]已被开发，旨在直接在VLSI硬件实现。主轴投影（PAP）技术[2]已被用于特征向量生成，它很好地代表了人类对图像相似性的感知，同时大大降低了维度。在基于PAP的矢量匹配算法中，我们引入了多分辨率和PAP核扫描的概念。结果，在灰度图像和二值图像中已经证明了非常鲁棒的图像识别。有趣的是，如图1所示，作为微型数字的集合而形成的大数字图案也可以不仅在组成小数字中而且在整个特征中被成功地识别。虽然本算法在计算上非常昂贵，但它已被设计为完全兼容于在我们为向量匹配[3-5]和PAP向量生成而开发的VLSI芯片上执行。
A human-like robust image recognition system inspired by a psychological brain model [1] has been developed aiming at direct implementation in the VLSI hardware. The Principal Axis Projection (PAP) technique [2] has been employed for feature vector generation, which very well represents the human-perception of similarity in images while substantially reducing the dimensionality. In this study, we have introduced the concept of multi resolution and PAP kernel scanning in the PAP based vector matching algorithm. As a result, very robust image recognition has been demonstrated in gray scale images as well as in binary images. Interestingly, a large digit pattern formed as an aggregation of miniature digits like that shown in Fig. 1 can also be successfully recognized not only in the constituent small digits but also in the entire feature. Although the present algorithm is computationally very expensive, it has been designed fully compatible to execution on VLSI chips which we have developed for vector matching [3–5] and PAP vector generation.