The Method And Its Research On Optics Disc Detection Based On Visual Saliency

Optic disc is an important component of retinal where vessels converge.In the diabetes diagnosis system,optic disc recognition plays a significant role in further diagnosis and analysis.Facing that current works on optic disc recognition usually perform weak accuracy,this thesis researches saliency detection methods to detect optic disc.The main work done in this thesis as follows:First,we propose a visual saliency detection method based on sparse reconstruction error.The image boundaries are extracted as background templates according to the assumption that salient objects hardly touch image boundaries.Based on the background templates,we construct a dictionary for each individual image and employ sparse reconstruction on the dictionary to obtain the reconstruction errors for each region to detect saliency area.Second,we propose a saliency detection method based on joint CRF and dictionary learning.This detection method segments the image into superpixels and represents each superpixel's appearance with sparse coding.Then we construct CRF to train saliency model.After that,we can get saliency area of test images.Finally,we propose an optic disc segmentation method based on Hough transform.We can get the shape detection result from saliency maps of bottom-up and top-down methods by using Hough transform,so that we can improve the detection accuracy and further optimize the result of the optic disc detection.We construct a new practical optic disc database to assist relative research works.