The Research And Implement On Identification Technique Of RBC And WBC Based On Intelligent Image Processing

The inspection of visible components in the urine is one of three greatest routine inspections. This detection item is of great value to the diagnosis, treatment and curative effect monitoring in the nephropathy,uropathy,hepatopathy and metabolic diseases。At the present time, doctors usually use manual checking by microscope in our country. The method results in two dominating abuses: firstly, different doctors get different results because the inspection of visible components is a demanding technology in the medical examination. Secondly, it spends too much time by manual checking so that the burden of the hospital is getting more and more weight year by year. So we need to product automatic analyzer to count visible components instead of the urine routine inspection. It is of great significance to improve the veracity of the diagnosis and to improve work efficiency and competing power of the hospital.The dissertation has analyzed, modified and also implemented by program many types of algorithms such as pretreatment of the urine image, Snake Algorithm used to extract the edge, the algorithm used to extract the feature and BP neural networks used to auto-recognize. My work is original and fruitful in the following algorithms:1. We have adopted a traditional algorithm, seed filling algorithm, to implement a few functions such as calculating the cancroids and the radius of the visible component and calculating the area, the mean gray value and the variance of gray value.2. The dissertation has expatiated the application of arithmetic operators of edge detection and a modified Hough transformation algorithm used in circle detection in the edge detection of the urine image and has proved the limitation of the two algorithms by experiments.3. The result of the experiment shows that it is very effective for the modified Snake algorithm to detect the edge of the visible component in the urine image. In additional, we also have discussed how to make certain the energy coefficient of Snake model.4. The calculation of the features has been implemented after the edge is detected by the algorithm of the Snake. The features include perimeter, area, seven step invariable moments based on edge, the length of the edge after fitting a straight line and the direction angle between two vectors.5. Red cell, white cell and crystal are recognized by BP neural networks and the veracity arrives at about 80%.6. All of the above algorithms have been implemented by c++ language in the Visual c++ platform.