Research Of The Key Techniques On Relic Image Classification System

This research paper is a part of the Research of Computer-aided Relic Recovering System supported by the National 863.Some key techniques and main algorithms of relic image classification system are presented. The system developed fulfills the retrieval and classification of culture relic images. And the main research is as the following four aspects:(1) Image PreprocessingAs the shooting problems and culture relic itself, the images captured often have many noises, so preprocessing of the images is necessary to improve the quality and to make the feature extraction reliable and exact. A simple and effective median smooth filter with boundary holding is applied to eliminate the noises. Meanwhile template smooth, grads sharpen and La-place sharpen methods are used to implement image's enhancement and to improve quality.(2) Representation of Image Features and Classification of Multi-featuresImage feature refers to some especially characteristic of image. Such as texture reflects some static rules of grey elements appearance, and shape describes the image's geometry property. In the paper, three kinds of image features included color, texture and shape are calculated and a method of multi-features used to classify relic images is investigated. Eight features of three kinds are selected to input to the inputting layer of neural network, and then the neural network is trained and tested toclassify the relic images.(3) Improvement of Zemike Moments and Image RetrievalAn improved algorithm of Zemike moment is proposed, and Zernike moment features are used to image retrieval. Experimental results show that the Zernike features can describe the region of the relic image and those moment descriptions are effective in image retrievaling.(4) Neural Network ClassifierAs a new pattern recognition method, the Artificial Neural Network (ANN) is applied in many fields, such as pattern classification, sound signal recognition and intelligent control and so on. The key algorithm of the neural network is deeply analyzed both with its excellence and disadvantages in this paper. Experimental results show that the improved neural network is an efficient classifier for relic images.(5) Relic Image Classification SystemBased on those algorithms above, I developed a relic image classification system and the main research objects are archaic china relic images of YaoZhou. A relic images database is established and the system can realize these operations such as image's preprocessing, feature's accounting and image's classification.