Research On Image De-noising Of Digital Radiography

With the progress of electronic and computer technology, modern medical imaging technology develops rapidly, digital process of the conventional X-ray photography equipment is carried forward at a high speed, and digital radiography (DR) also emerges with this opportunity. DR, which transforms X-ray signals into digital images via the charge-coupled device detector (CCD) or a flat panel detector, has many advantages, such as fast imaging speed, low dosage of X-ray required, high image resolution, wide dynamic range, good imaging continuity, easy to operate and so on. As a result, it is widely used in clinical diagnosis, making it one of the most popular large medical equipments.However, in the process of DR imaging, a great variety of complicated noise is brought about due to various reasons, reducing the quality of DR images and interfering with the doctors'clinical diagnosis. For this reason, researchers in academia and industry have studied on the problem of DR image de-noising and have made a series of achievements. But because of its simplicity and roughness or weak pertinence to DR image, the traditional DR image de-noising methods such as linear filtering can not easily balance between the effective de-noising and retention of more detailed information.In order to explore new methods to eliminate the noise in DR images, this thesis bases its study on familiarity with and mastery of the principle of DR imaging, then it presents a thorough study of DR image de-noising, analyzes the noise sources and characteristics, determines the main noise for study, and researches on distinct de-noising methods for different noise respectively. To sum up, this thesis completes the following research work:(1)For impulsive noise in DR images, this thesis first analyzes the generation and characteristics of impulsive noise, then summarizes and analyzes the traditional de-noising methods of impulsive noise. After that, a new switching median filtering method is introduced in particular, which establishes an accurate noise hybrid model based on the statistical property of impulsive noise. Finally, this thesis analyzes and discusses the introduced method.(2)In order to remove Poisson noise in DR images, this thesis studies the threshold de-noising methods in multiwavelet domain and improves the traditional thresholds and threshold functions using the covariance shrink method. The covariance shrink method in this thesis has a close relation to Poisson noise, and it can calculate different shrinkage thresholds for each high-frequency detailed multiwavelet subbands, showing good adaptability.(3)For Gaussian noise in DR images, this thesis studies the improved contourlet thresholding. Firstly, contourlet transform combined with recursive cycle spinning is introduced to decompose the image, then the traditional threshold and threshold function are improved, and finally accelerated computing algorithm based on GPU is carried out.