Research Of Pattern Recognition Based On Cognitive Modeling Of Primitive Topology CGA Modeling

Since we came to the new century, our human have faced various information ofdigital multimedia, image is the most typical miniature of digitized information, it hastransport many valued message, and it is one of main source for individuals to enquirethe outside knowledge. There are many kind of images, we can see them in a lot oftransmission medium, icon and word image are the focus research objects of imageretrieval based on content. This paper simulating human cognitive pattern recognitionmechanism, proposed a feature extraction method based on the reading cognitive model,extracted image feature based on visual information, and put forward a universalpatterns recognition method based on primitive topology modeling. The ArtificialNeural Networks(ANN) was used to model the primitive of the pattern as its bigcomputing ability, and the Conformal Geometric Algebra(CGA) was used to model therelationship among the images primitive, furthermore, the Hidden MarkovModel(HMM) was used to model the image overall structure as its strong merit ofdispose time series data.Based on the image recognition method which this paper proposed, the icon andword image were seen as our experiment object. We used the sliding process of slidewindow to imitate the moving process of center of the retina when human look at image,in this process, we completed the transform of2D image to1D feature vector. Therecognition model took the hybrid of ANN and HMM based on CGA to achieve imagemodeling. CGA has the ability of convert geometric problem to algebraic problem, so itis used to transfer the modeling of conformal geometry part of images to the algebraicdescription of the clustering problem. ANN has several merits, such as self-organization,self-adaption, computing and anti-jamming. HMM has the merit of dispose time seriesdata, so the combination of ANN and HMM based on CGA modeling the relationshipamong the primitive of images is very proper for this paper proposed image recognitionmethod.In order to verify the effectiveness and universality of the proposed method, thesimple structure icon and complex structure word image were used to be the experimentobjects. According to the results of the experiments, for the different kind of image, theproposed recognition method not only give the good recognition results, but also reflectthe computer imitate the mechanism of human recognize image. Furthermore, when theexperiment material has some information missing, the recognition results was alsogood, which shown the method’s anti-jamming deformation ability and robustness.