Research On Capsule Endoscope Image Recognition

Digestion subject’s an important milestone is capsule endoscope technology, which is combination of micro-electronic technology, image processing technology, wireless communication technology and biological medicine. It Has now become a important examination methods of digestive tract disease inspection, however capsule endoscope image number is huge, as the rate of only two photos per second, inspection process of six to eight hours produces images of43200-57600, most of which are normal images and repeated images, abnormal images account for small part, and now film frame rate of hardware technology has increased to8frames per second, a patient’s image number reaches to hundreds of thousands, it takes a great working and spends a lot of time for doctor to make a correct diagnosis through check all of these images one by one, what is more, total visual check is easy to weary and miss valuable diagnostic information, the diagnostic efficiency is extremely low, these problems limit capsule endoscope’s widespread use. Therefore, capsule endoscope image’s visual check has become bottleneck of restricting its development.If we can identify diseased images from the patient’s mass images, this will no doubt reduce the doctor examine time and help clinical doctors screen of diseased images rapidly. Therefore, image analysis and retrieval technology has been applied to the medical field. Because of the particularity of the capsule endoscope image, it is still an unresolved problem that how to organically integrate image retrieval technology with capsule endoscope image so as to provides physicians with a favorable and convenient tool to retrieve diseased images. Combining with color capsule endoscope image characteristics, we put forward the content-based image retrieval (CBIR) technology to wireless capsule endoscope images’intelligent recognition, this paper studies on how to identify abnormal images from a patient’s mass images. Research indicates that retrieval results are very different between RGB and HSI space, the reason is that the capsule endoscope images have rich colors, capsule endoscope image expressed in HSI space is in line with the doctor’s vision. We extract color and texture feature of capsule endoscope image’s HSI component, because of the capsule endoscope many lesions image containing the characteristics of red bleeding, which indicates that the H (hue) component plays important role during diagnosis, during the image similarity measurement, we set up different weight to HSI component but no average weight distribution, the simulation results show that the proposed method in this paper obtains better retrieval results than other methods. Through comparing the patient’s each image to each other in the standard library standards cases, choose the maximum similarity as its index values, the patient’s images output from the high end, which ensures that suspected abnormal images can also be timely and effectively detected even if a patient contains a variety of digestive disease, not limited to a disease, it is very meaningful for the clinical inspection, clinical doctors can make a diagnosis through just check the small part of the patient’s mass images. Compared to check40000～50000images, this not only greatly ease doctor’s workload but also improve the diagnosis accuracy rate. It will be more in line with doctor’s visual characteristics by changing weights of HSI components according to experience.Combining with the modern fuzzy mathematics and artificial neural network, this paper put forward wireless capsule endoscope images automatic recognition based on back-propagating (BP) neural network. According to the capsule endoscope neoplasm images and normal image features, this paper applied image recognition algorithm to capsule endoscope image identification field, the paper proposed that taking FTS (fuzzy texture spectrum) statistical characteristics of six component as capsule endoscope image texture features, adopting the BP neural network classifier to identify, multiply BP vote for the final decision of the results. The program were coded by matlab and C++programming language, the experimental results indicate that the algorithm is rational and feasible in capsule endoscope image recognition, which can assist doctor screen of abnormal images from a patient’s mass images.