Research On The Detection Technology Of Calligraphy And Painting Based On Pattern Recognition

With the market of painting and calligraphy is enlarging,all kinds of painting and calligraphy counterfeiting means is becoming increasingly sophisticated,rampant counterfeit has become a malignant tumor of painting and calligraphy market,so the identify the authenticity of painting and calligraphy also appears increasingly important.At present,for the identification of painting and calligraphy works,scientific and technological means intervention is relatively small,mainly based on the experience of the connoisseurs.However,the biggest drawback of this method is that the effect of the subjective factors of connoisseurs,the lack of objective and quantifiable indicators,so the authenticity of the work entirely depends on the subjective feelings of the experts.With the development of pattern recognition and computer vision,it has become possible to identify the authenticity of calligraphy and painting with the aid of computer.It can overcome the disadvantages of the traditional identification method.Therefore,this paper presents a study based on pattern recognition technology of calligraphy and painting authenticity identification,which aims to overcome the limitations and difficulties which the traditional painting identification technology faced with.In this paper based on the research and study of the pattern recognition technology,this paper analyzes the shortcomings of the existing algorithms in the application of the painting and calligraphy analysis,and puts forward some improved algorithm.The main work of this paper is as follows:As the sample images will be introduced in artificial error,for example,translation,rotation,scaling and the existence of visual angle translation,image blur,noise and other factors,and the traditional image registration technology is difficult to handle so many cases at the same time.In this paper,an image registration process based on SIFT(feature transform Scale-invariant)scale invariant feature transform algorithm is proposed.The registration process can resist many changes mentioned above.Because of the high similarity between the images to be identified,so the general and traditional descriptors cannot be effectively identified.Therefore,in this paper,a lot of texture descriptors and gradient based descriptors are studied.Different feature descriptors are used to extract feature vector according to different painting type,in order to achieve the correct identification.The last part is to work on the feature classification.For the identification of seal,the experimental sample is easy to get,so this paper proposes a multi classifier fusion strategy based on machine learning.SVM(support vector machine),KNN(k-nearest neighbor),and NBC(Naive Bayes Classifier)are used to classify the feature vector and the results are combined by a fusion criterion.Finally,we get an average of 96.75%accuracy.But for the calligraphy and painting,experimental sample is relatively rare,so this paper makes a judgment on the characteristic data through some feature measurement and statistical analysis.And the effectiveness of the algorithm is proved by a lot of experiments.