Vein Image Enhancement And Its Implementation To DSP System

In venous puncture activities, the patients’ blood vessels may not be so obvious for nurses to find, especially in the cases of children, women, obesity or those with dark skin, which makes venous puncture difficult. It is because vessel diameter is vary thin or cotex is such thick cortex. So it is very important for nurses to find the blood vessel in patients quickly and accurately. Nowadays, the mainstream way to obtain venous angiography images is by using the infrared reflectance model, which has the better environmental adaptability. But duing to poor infrared filters and reflected infrared light wavelength limits, most of the original images is poor in quality, with low contrast and much noise. It makes direct venous enhancement difficult.This paper refers to the mainstream algorithm, such as the image contrast lifting algorithm and image texture extraction algorithm by using directions, and designs a new algorithm model. This model can not only make up for the weakness of the histogram equalization algorithm or retinex algorithm which cannot enhance weak contrast vessel images, but also fix the disadvantages of algorithms like Gabor transform, direction filter of feature extraction or Hessian matrix method, which cannot be put into real-time processing applications for their poor algorithmic complexity.The core idea is to put Hessian matrix method and direction filter in reference, considering the neighborhood direction for each pixel in the image. By using second derivative Haar-Like gradient filter template, it filters the images with multiple directions at the same time (through code optimization, it can make the DSP’s pipeline parallelly run), and summarizes all the weighted directional filter arithmetic results to extract the center line features of vessels. That is, it ultilizes mean filter instead of the Gauss filter to extract texture features. On the other hand, it can also filter out background noises. Besides the new method can alse eliminate noises caused by skin scaly reflected light by using first-order derivative Haar-Like filter template. The classic method is to gain the pixel pattern directions by direction filters and or Hessian matrix and then ultilize a appropriate template to extract vessel texture features. The method in this paper combines the two steps of classic into one step, and at the same time, makes full use of filtering template’s to eliminate the noises near the vessel boundaries.In order to improve the accuracy of image texture feature, the method in the paper conducts a preprocessing operation before using the Haar-Like filter. The project ultilizes image histogram to make the pixels normalized, and increases the dynamic range of pixel gray level, which will increase the contrast of the original image. At last, the algorithm is transplanted to DM6437hardware platform By optimize the algorithm, it can achieve to real-time enhancement of vein texture.