Research On Segmentation Of Blood Vessels In Fundus Images

Image processing and analysis techniques have been widely used in the field of medical images recently. Due to the advanced computer technologies, medical images significantly improved the clinical diagnosis and treatment. The changes of its distribution structure and morphological characteristics could reflect the severity of diseases to a certain extent. However, as the restriction of imaging conditions and physiological structures, the complex features of fundus images bring great difficulties to the segmention of blood vessels.In this paper, the segmention technique of retinal blood vessels has been studied and a method based on discret Guassian convolution kernel and Hessian matrix is proposed. This method firstly enhanced the characteristics of blood vessels by pre-process. Then a retinal vascular model is established for anlysis, and the discret Guassian convolution kernel is given to detect centerlines of blood vessels. The sub-pix position of centerlines is calculated through Hessian matrix and then connected with a certain connection algorithm. The vascular boundary can be obtained by two method based on gradient and vascular centerlines. Image characteristic parameters measurement methods are also studied in this paper. The normalization dimension is obtained by an image calibration method firstly, and then the vascular parameters are measured by two methods based on manual and vascular segmentation. An optic disk localization algorithm based on the Canny operator and centroid detection is put forward in this paper. On the basis of analysis and research, retinal image processing system with retinal vascular segmention and parameter measurement modules are realized in C + + Builder development platform, combining the background of fundus camera project.This paper realized the precise segmentation of retinal vessels with subpixel level positioning accuracy, which effectively meets the requirement of medical research and clinical diagnosis. The proposed vascular segmentation algorithm is tested utilizing two international public fundus images datebases. The results are compared with manually segmentation images by two defined evaluation standard "accuracy" and "contactratio". Experiments prove the effectiveness and accuracy of the algorithm.