Research On Paper Testing Technology Based On Image Processing

With the continuous development of image recognition technology,more and more image processing methods have been widely used in medical research and clinical practice.Image recognition technology has become more and more important in medical images.In this paper,we main study the colloidal gold protein test strip image which is widely used at present.Through the method of image processing,the image features of the test paper are extracted,and then the paper information is identified by classifier.The core part includes the location segmentation and information identification of the paper image.The rapid and accurate classification of test strip images is done by computer to resolve the inefficiencies of the human eye and the disadvantages of current related identification instruments.Researches and comparisons are made on commonly used localization algorithms,and several image preprocessing algorithms for test strip images are proposed,such as image scaling,image graying,image binarization,and image edge detection.According to the characteristics of test strip image,a tilt correction method for test strip image is proposed.The edge correction method combined with Hough transform can better realize the tilt correction of the strip image.With the help of localization algorithm combined with edge detection of the test strip boundary,the area where the test strip reaction is located.The pixel color accumulation method and projection method are used to achieve the segmentation of the test paper color area,and further the test paper color area is extracted from the test strip image,and use it as a sample image to be identified.The stripe image histogram,gradient histogram,image correlation,inertia and the moment of deficit,etc.are used as the texture features of the test strip.At the same time,the hue,saturation,and lightness of the image are used as the color of the test strip image in the HSV color space.Finally,the 21-dimensional feature vector is obtained as the input of the classifier.After the feature vector of the sample is extracted,the training set sample is selected to train the classifier formed by the SVM and the BP neural network.We can get different classifier models to achieve four types of test strip sample classification.In order to achieve the optimal classification result of SVM,this research finds the best parameters corresponding to the model through the grid parameter search method,and then analyzes different kernel functions with different numbers of samples to determine the selection of SVM kernel function.In the BP neural network,we aim at the problem that different hidden layers cannot be accurately selected,experiments are performed on the same test set with different hidden layers to find the optimal corresponding hidden layer layers.The final experimental results show that the average classification accuracy of the two classifiers for the four types of test samples is about 90%.The two classification methods achieve relative good classification results for the test strip images.The results show that the proposed method based on image processing can identify and classify the test strip images well.For the strips of same characteristics but different types they can satisfy the detection requirements,which has a good practical value in practical applications and development prospects.