Research On The Feature Extraction And Recognition Algorithms Of Paper-cutting

Paper cutting is a long history of traditional Chinese folk art, with the continuous development of China's animation industry, paper-cut works will be a very good cartoon material. To achieve this art form in digital media, first convert the paper-cut works to digital images stored by computer, generate further work, paper cutting patterns in these processes have a very important role, which determines the final paper-cut images forms. The different Paper-cutting patterns'precision classification and recognition is the basis of image applications,This paper studied the paper-cutting patterns'identification and submit some recognition algorithms,.experiments show that a relatively good recognition performance could be achieved to recognize paper cut-out patterns of exaggerative deformations.Paper works primarily from the following aspects:(1) This paper studied the characteristics of paper-cut images and created paper-cut patterns library. Paper-cut patterns on the collected images were preprocessed to eliminate noise on the image and segmen the isolated individual patternts from the complex paper-cut images. Through analysis of paper cutting patterns'characteristics, set up a containing 63 training samples and 350 test samples paper cutting patterns libraries, these patterns cover the creation basic patterns, prepare for the subsequent identification.(2) A method of Paper Cut-Out patterns recognition based on moment invariants and geometric characteristics was proposed。The traditional seven invariant moments were used as feature vectors, with translation, rotation and scale invariance.At the same time the six geometric invariant features were extracted from pattern image.The seven moment invariants and six geometric invariant features normalized are input into BP neural network based on Levenberg-Marquardt algorithm and the trained network are used to be a classifier to realizes the image pattern recognition.Experiments proved that this method have a good performance in recognizing paper cut-cut patterns of exaggerative deformations.(3) The wavelet analysis and other feature extraction methods were combined. Wavelet analysis have the Multi-resolution features, if feature vectors are extracted based on wavelet multi-resolution, it will enhance the feature vector's ability to represent the images. This paper extracted wavelet energy feature, singular value feature, NMI features and wavelet moment based on the wavelet analysis respectively, and do a comparison recognition test. The wavelet transform was applied to extract high-frequency energy components expressed in different directions as the final feature vector from the paper-cut patterns images。In the SVD and NMI medthods,the paper cut-out image normalizedand and binaryzation are discomposed in low frequency component by tow-dimension discrete wavelet,then distill the singular value and NMI from low frequency component, after reducing dimension processing and making unitary dealing we get the characteristic vector of the image. This combined approach effectively uses of wavelet multi-resolution features, eliminating noise interference, and remain the SVD and the NMI feature of its own characteristics,and the method is simple, easy to achieve.Wavelet moment feature of image can reflect the image's part and whole characters and has strong anti-jamming ability. Combined with the characteristics of paper-cut images, the different mean and standard deviation of eigenvector were used to achieve N class model feature selection. Experiments show that this method can effectively remove noise, better identification of exaggeration deformation patterns.Experimental results demonstrate that the above method can well identify paper cutting pattern image of exaggerated deformation.The algorithm's complexity are also able to meet the computer real-time requirements as a basis for the next paper-cut images automatic generation.