Study On Image Segmentation Technique Of Touching Rice Based On Watershed And Pits

Food is the lifeblood of the country,food security has been elevated to the national strategic level.Rice is the main varieties of grain,digital image processing technology is an important means in food quality testing.For the counting and feature extraction of large number of touching grain,segmentation technique for touching is required. In recent years,segmentation technique for touching got continuous development and it is widely used in medicine,industry and agriculture.The main purpose of this paper is to find an improved method for touching grain segmentation which will separate the rice and make the results more accurate and features more completely.Traditional watershed algorithm is sensitive to noise, easy to form over-segmentation.In this paper,over-segmentation is reduced by reducing the number of the minimum point in the gradient image of the rice image. The specific process is this, firstly reducing some pseudo minimum point by using the extended minimum transform with adaptive H,then merging the remaining minimum point by dilation and erode operation which will make a rice granule has only one minimum point and minimum point of the touching rice are not connected,finally segment the modified mark image by marked watershed algorithm. Experiments show that the touching rice have been separated and the over-segmentation phenomenon has been suppressed, the effect is well.For the problem that the boundary of the rice is not accurate after the morphological operations as the watershed method,besides a relatively accurate boundary shape feature is wanted.This paper implemented a new segmentation method with pits, firstly finding the pits by combine the method of five-point angle and the method of chain code difference,then divided the image according to the number of pits.This paper solved the precision problem of finding pits in the segmentation which only using the method of five-point angle. Experiments show that the position of the pits is accurate and the segmentation effect is well.