Research On Classification And Detection Of Stained Coins

At present,as one of the most important currencies in circulation,coins play an auxiliary role in the production and circulation of social commodities.With the long-term use of coins,it is inevitable that the coins will be polluted and damaged in different degrees,which will make the stained coins not accepted by the public in the circulation link and affect the normal use of coins.There is a contradiction between a large number of coins in circulation and a small number of detection workers in the classification and detection of stained coins by artificial method.At the same time,due to the different standards of artificial identification of stained coins,the results will have the human subjectivity.In order to save the cost of classification and detection of stained coins,improve the efficiency of classification and detection of stained coins,and eliminate the errors caused by human subjectivity,it is a trend to use intelligent equipment instead of artificial detection to finish the classification and detection of stained coins.Based on the image processing technology,this paper studies the classification and the detection of stained coins.On the classification of new and old coins,this paper proposes a classification method of new and old coins based on regional statistical features;On the detection of stained coins,this paper proposes a new detection method of stained coins based on the improved SIFT.The main research contents are as follows:(1)This paper studies the classification of new and old coins.In order to ensure the invariance of translation,scale and rotation,this paper extract the statistical features of the coin image to classify the new and old coins.At the same time,in order to preserve the spatial structure information of the features,this paper proposes a classification method of coin image based on regional statistical features.In the training stage,firstly,extract the coin circle pattern from each image in the training sample;Secondly,normalize the circle pattern;Finally,divide the coin circle pattern into several Annular regions and one circle region.Extract statistical features in each region as the features ofcoins.Using the statistical characteristics of the training samples to train classifier.In the classification stage,for the input image,extract the statistical features according to similar steps,and then use classifier to classify the old and new coins.Experiment with the new and old coins of one yuan,The experimental results show that the new and old coins classification method proposed in this paper can get higher classification accuracy and speed.(2)This paper studies the detection method of the stained coins.Considering that there is a big difference between the pixel of the stained position of the stained coin image and the pixel of the corresponding position of the new coin image,this paper directly compares the corrected stained coin image and the new coin image to detect the stain.In this paper,use SIFT to correct the stained coin image.In view of the large feature data and high time complexity of SIFT,this paper proposes an efficient method for SIFT feature extraction.In a certain scenario,if the scale of coins in the image is fixed,the extracted SIFT features should be distributed in a specific scale space.By extracting SIFT features from different coins and according to the distribution of correctly matched feature points in scale space,dimensionally reduce the scale space constructed by SIFT algorithm.The scale space after "dimension reduction" only includes the scale space which contains most of the correctly matched feature points.Therefore,it can avoid extracting a large number of redundant feature points,effectively reduce the computational complexity of SIFT,and further improve the detection speed of the scale of coin.Based on the improved SIFT algorithm to detect the stained one yuan coin images,the experimental results show that the image scale space which originally contains 35 layers is reduced to a scale space which only contains 4layers for the detection of stained coins.Compared with the classical SIFT,the improved SIFT proposed in this paper improves the calculation speed by two orders of magnitude,and can effectively achieve the detection of stained coins.This paper mainly studies the classification of new and old coins and the detection method of stained coins.According to the classification of new and old coins,this paperproposes a new and old coins classification method based on the regional statistical characteristics.we use the classification accuracy and speed to evaluate this classification method.The experimental results show that the results of the new and old coin classification method proposed in this paper are satisfied.In this paper,a new method based on improved SIFT is proposed to detect stained coins.This paper uses the number of feature points,the logarithm of total matching feature points,the logarithm of correct matching feature points,the matching accuracy of feature points and the matching time to compare this method with classical SIFT.The experimental results show that the improved SIFT can reduce the extraction of redundant feature points,and can effectively achieve the detection of coin contamination.