Research On The Microaneurysm Detection Algorithm Using Weighted Naive Bayes Classifier And Radon Transform

With the development of computer technology,the digital retinal image processing and analysis techniques have become more mature.Diabetic retinopathy is a common and severe complication of diabetes and it is the cause of new-onset blindness.Currently,the major problem in the treatment process of diabetes is how to safely and efficiently perform digital retinal image screening and timely identification of diabetic retinopathy as early as possible in order to take appropriate measures to prevent blindness.This thesis mainly studies on diabetic retinopathy in the earliest sign called microaneurysm.This thesis proposed microaneurysm detection algorithms based on weighted naive Bayes classifier and Radon transform.This thesis mainly researches on the feature extraction in the radon transform domain.The essence of the algorithm is to detect the significant linear structure in an image.The main contribution of the transform is that it converts the detection of a single point to the detection of linear structure.The radon transform used in this thesis is defined by the the Cartesian coordinate system called slant stacking.It needs two domains to properly address all directions using slant stacking.There is a common problem,which is the variable length of the integration line.This bias can be reduced by dividing the integral of each line by its length,which corresponds to the calculation of the mean of the intensities along the integration line.The features extracted in this method will be used in the classifier for the location of microaneurysms.Finally,this thesis adapts the features extracted in the radon transform domain and image domain to the weighted naive bayes classifier so as to locate the real microaneurysms.This thesis chooses the naive bayes classification algorithm.Bayes algorithm has strong robustness of the model,but it assumes that each component used in the classifier has the same influence on the result.To solve this problem,this thesis adopts the weighted naive bayes classifier,in another word,assigning an appropriate weight to each feature,which indicates the degree of influence on the results.Therefore,another question about the algorithm is to determine the weight coefficients.This thesis uses the genetic algorithm.From the experiment result we can see that the sensitivity of the detection can be significantly improved and the average false positive rate can be lower.Therefore,the proposed algorithm is effective.