Research On The Method Of Urinary Sediment Image Recognition Based On Multiscale Computation

The urine analysis is one of the convention examination project in today’s hospitalclinical examinations. The urinary sediment image analysis provides the identification andcounting results about red cells, white cells and other components in urinary sediment tothe doctors. It helps the doctors to determine whether patients have urinary disease, kidneydisease and other diseases. Otherwise the urinary sediment inspection has great clinicalsignificance for disease diagnos. The ingredients in urinary sediment image is various andcomplex. So traditional human eye test is hard to meet the requirements of clinicaldiagnosis because the heavy workload and artificial error rate. With the rapid developmentof computer technology and digital image processing, the computer-assisted automaticclassification and recognition of urinary sediment images comes into being. How toimprove the recognition accuracy rate and recognition speed is becoming a research hotspot.Through the research, summary and a lot of experiments on the traditional imagerecognition algorithm, we find that the traditional algorithms have some disadvantages atcalculation and computational complexity. This article puts forward a new rapid urinesediment cells location and classification algorithm.The article has mainly carried on the research and analysis from the following severalaspects:(1) A new fast location algorithm based on multiscale is presented for the complicatedcomputing amount problem in multiscale in the traditional image location method. The fastlocation algorithm which combines the Viola integral image and multiscale imagemoments scans on the image to realize fast calculation. The algorithm has manycomputation properties such as automatic filtering, the same amount of computing undermulti-scale and higher computing speed etc... This paper discusses the principle, thefeatures selection and application of this algorithm.(2) In this paper the new fast location algorithm based on multiscale is mainly appliedin the location of urinary sediment image cell. Using of the Matlab software to experiment on the cell of urine sediment image. This new algorithm scans on the image fast in order toobtain the characteristic image to realize the edge detection. And then, the unique cellcenter is obtained by the fast computation, converged and enumerate on target moment ofimage.(3) A new pattern recognition method based on the gene regulatory network andBayesian probability theory is introduced in this paper, which is called support featurenetwork (SFN) method.This method realizes classification by finding characteristicsnetwork in different urinary sediment to implement the structural description. And thedescription of structure is used to build corresponding classifier of the classification.Finally, apply SFN in the recognition of urinary sediment. The experimental results showthat the red blood cell, white blood cell and crystal can be classified and recognisedsuccessfully.