A Study To Wavelet－Based Computerlzed Radiography Image Filtering Methods

Wlththe fleetingprogress ofcomputerteccioanddlgltaltechology,newly devefoped duntlzed medicine techniques go on expending,while computedradlographyls atyplcalmethods．itheprocedure ofcomputedradlographylmaging,we should irstlygetacross the characters ofklnds ofnolses andthe relationshipbetweenthe image signals andnolses．Basedonthe specialties o土computedradlo－graPhy images and medical image processing,we h。e study the filtering methodsr competed radlography images noises．our mMor wo＊Indudes:［1］ On the base of analyzing computed radlography imaging system In detail,the author th让出 that the mnyor… noises are Gausslanwie noise and Polsson noise． Then,the dlfferemrelationshlp ofbe加eentwo kinds ofnolses and sipal were studied completely．［21 By conslderingboththe charactenstlcs ofcomputed radlography images andthe statistical feres ofwavelettransformed images,a multlscale image iltenng algonthln,which based on wo－state hidden markov model(HMM)and mlxtUle Gausslan statistical model,has been used to decrease the Gausslan white noise In compmed images．We expenmems as well as the comparison ith other denolsing methods werepresentedatlast,[3] Because of both the specialties of Poisson noise and characteristics of wavelettransformed of Poisson data, the Harr wavelet, which is a tool to transfOrm theimages, and mixture Beta statistical model, Which described the numerical relation-ship between adjacent scales of the transformed Poisson data, were adopted. Thenwe set up the hidden markov model on which the Poisson noise was filtered acrossscales according to the upward-downward algorithm. The perfOrmance of the newfilter is accessed wtth simulated data experiments and tested with actual computedradiography images.