Image Feature Recognition Of Ancient Ceramic Shapes And Ornamentation Based On Artificial Intelligence

Ancient ceramics are a comprehensive reflection of the development of scientific and technological levels,artistic styles,and cultural elements throughout history.In particular,Chinese ancient ceramics,via their exquisite craftsmanship and rich shapes and ornamentations,fully demonstrate the essence of traditional Chinese culture and have become representative of national culture.Therefore,they have important historical,artistic,and scientific research value,as well as collection and investment value.Today,the authenticity of identification of heritage is suffering an unprecedented crisis of confidence.There are some limitations and discrepancies in the existing methods of ancient ceramics visual identification and science and technology identification.Features such as the shape and ornamentation are not quantifiable via visual identification,and the conclusions are too subjective.Due to the selection of raw materials according to local conditions,different molding and decoration processes,and different firing temperature,a wealth of image features of utensils and patterns are created.Features such as the shape and ornamentation are not quantifiable via visual identification,and the conclusions are too subjective.In this study,the main visual features of ancient ceramics,such as shape,pattern and inscription recognition,are extracted,quantified and identified by machines,and the idea and implementation method of nondestructive intelligent recognition of ancient ceramics by machine replacement experts are discussed.The results were as follows.(1)The image features of ancient ceramics are closely related to the selection of specific raw materials and process technology.The complete acquisition of image features of ancient ceramics is the premise of realizing the image feature recognition of ancient ceramics.The extraction quality of region growing-based extraction method is closely related to the background pixels,and the method is not generalizable.This paper proposes an extraction method based on deep learning.Take easyDL as the support platform for deep learning,training on a total of 5834 sheets,272 categories of ancient ceramic image features were extracted from blue-and-white porcelain and Yaozhou and Yue kilns,after extraction and verification,the results demonstrate that the average complete extraction rate was greater than 99%.The implementation of deep learning method is summarized and compared with the traditional region growth extraction method.The method based on deep learning can robustly change the extraction quality with the increase of the learning amount and is generalizable.It is a new method for feature extraction of ancient ceramic image.(2)Due to the different raw materials and techniques of ancient ceramics,there are rich characteristics of the wares.In order to realize the recognition of multi type image features,this paper proposes a specific method of ancient ceramics shape feature extraction and recognition based on binarization(Otsu binarization algorithm),morphological processing(corrosion,expansion,watershed segmentation),eight-chain code consistency detection,and similarity detection.Shape feature extraction and recognition verification of the proposed method was conducted with Mei and Dan vases,the final recognition rate was 97.12%and 98.36%.Twenty-six images of ancient ceramics were selected for batch shape feature extraction and recognition,and included blue-and-white porcelain and Yaozhou and Yue kilns.The average recognition rate of 24 samples is higher than 90%.The features of the remaining two samples could not be extracted and recognized due to their high complexity and image quality problems.The results show proposed method is a new effective method for shape feature extraction and recognition of ancient ceramic images.(3)Due to the different glaze selection,trace element content,decoration and firing process,the image features of ancient ceramic decoration are complicated,and there are rich color and texture features in it,which show a variety of characteristics under the machine and human vision.In order to realize the image feature recognition of ancient ceramic decoration,a multi-dimensional feature recognition method is proposed in this paper,which combines the gray level co-occurrence matrix(machine vision)and Tamura texture(human vision)features,the average Euclidean distance is the recognition result,the color histogram(machine vision)and HSV color space(human vision)features are mixed,and the average similarity is the recognition result.Four samples from Yaozhou and Yue kilns including celadon ceramics,pastel,and enamel characteristics divided were into two cases,namely those with salt and pepper noise and those with partial loss.The results demonstrate that the HSV features and color histogram features changed notably after 20%salt and pepper noise was added,and the recognition rate was less than 80%.Additionally,the Euclidean distance between the gray-level co-occurrence matrix and the Tamura texture feature vector changed greatly.After random partial loss,the HSV feature and color histogram of the color-space features changed only slightly,and the recognition rate was greater than 90%.The Euclidean distance between the gray-level co-occurrence matrix and the Tamura texture feature vector also changed only slightly.The results show that the proposed method can effectively realize the quantitative recognition of the image characteristics of the ancient ceramic ornamentation in different ages and kilns.(4)The visual characteristics of ancient ceramic inscriptions are closely related to the selection of raw materials,element content and decoration technology.Due to the progress of science and technology,there are rich types of inscriptions.In order to effectively identify the image features of inscriptions,this paper presents a classification recognition method of inscription image recognition based on deep learning(unsupervised learning)is proposed.The EasyDL platform is used to support the Ming and Qing Dynasties' two dynasties official kiln inscriptions,the local image features are enhanced locally after training and learning,and the training results are verified.Among the 540 inscriptions of 12 "Ming Xuande" and 1200 pieces of 20 types of inscriptions of the Ming and Qing Dynasties categories,the highest recognition rate was close to 100%,the lowest recognition rate was higher than 97%,and the average recognition rate was higher than 99%.Compared with the recognition method of single Chinese character after the segmentation of inscription image recognition,this method can effectively and comprehensively recognize the features of ancient ceramic inscription image.(5)Ancient ceramics are a kind of special material products(works).According to the selection of raw materials and different technologies,a variety of image features are produced.In order to realize the recognition of ancient ceramic image features.It is based on the methods of feature extraction of the whole ancient ceramic image,shape image feature extraction and recognition,multi-dimensional feature mixed ornamentation image feature quantification recognition and feature classification recognition of inscription recognition image based on deep learning,supported by a three-tier B/S web application system and cross-platform system language calls.Take database services,deep learning packaging,and third-party digital image processing as specific implementation methods.The service layer fusion and relearning mechanism is put forward,and the intelligent recognition system of ancient ceramic shape and ornamentation images is preliminarily realized.The results of the verification test are in line with the expectation,which verifies the effectiveness of the recognition system of ancient ceramic pattern and decorative patterns.