Research On Image Registration And Its Applications In Astronomical And Microscopic Image

Astronomical and microscopic images are two special fields in image processing field.Recently,with the ever-changing development of astrophotography and digital microscope system at home and abroad,the demand of new astronomical and microscopic image processing techniques and applications are urgent.According to these needs,this thesis researches on image registration and its applications in astronomical and microscopic image.In astronomy field,image registration is mainly used to correct the translation and rotation caused by the devices' vibration and the Earth's rotation,scale,and non-rigid deformation aroused by the atmosphere turbulence.In microscope field,image registration is prone to be used for assisting multi-focus image fusion to extend the depth-of-field or image stitching to broaden the field of view.In the application of deep-sky image live stacking,this thesis analyzed the defects arousedby applying the prevalent feature-based methods in deep-sky image registration and proposed three different deep-sky image feasible live stacking algorithms based on the characteristic of deep-sky images.Firstly,a thresholding method based on Otsu and centralization is proposed to implement star detection and a creative descriptor based on star's property and geometrical information is put forward to describe stars.The introduced idea of binalization and the simple and effective mismatching rejection scheme solved the problems encountered by the existing deep-sky image registration methods,such as the low efficiency and high resource utilization.Secondly,the idea estimating the dominant orientation by the nearest stable stars creatively solved the hard problem of dominant orientation assignment suffered by the conventional descriptors.The estimated dominant orientation is more reliable and stable than that done by the local statistic information.Compared with traditional triangle-based methods,without constructing and saving amount of triangles or polygons,the proposed descriptor improved the efficiency and reduced the resource utilization greatly.The firstly proposed local path adaptively choosing scheme effectively avoided the defect of determining the local patch size empirically.The first attemptation of embedding star's property and spatial geometrical relationship into SURF descriptor improved the robustness of SURF.As a new descriptor,the proposed descriptor shows better rotation invariant and noise resistance.Finally,the improved efficient triangle-based method creatively proposed a stable star division method by analyzing the single pixel intensity and star size histograms of detected stars,and based on the analysis the triangle construction progress is optimized.The adoption of the similarity degree successively solved the fewer correct matches and lower recall problems caused by reducing triangles.The proposed similarity-threshold automatically determined method solved the hardness of choosing a correct threshold.Experimental results show that the proposed method greatly enhanced the star matching efficiency,reduced the resources requirement and guaranteed a high level of matching precision.In planetary image live registration,planetary image easily suffered from the devices'vibration and the Earth's rotation,this thesis performed deep researches on phase correlation because of the fact that planetary images lack enough texture features to be extracted.A sub-pixel registration method was extended to solve the serious efficiency reduction during processing high-resolution images and achieved excellent live registration performance.For the non-rigid deformation caused by atmosphere turbulence,this thesis researched on the application of non-rigid registration in atmosphere turbulence removal.According to deficiency of the existing techniques and the characteristics of planetary images,a feature and B-Spline based free form deformation(FFD)registration was proposed to implement planetary images atmosphere turbulence removal.The proposed method creatively described the sparsely and densely sampled grid points using SURF descriptor for reference and deformed images respectively.According to the minor and local specialty of the non-rigid deformation in planetary images,a spatial constraint was embedded into feature matching and that boosted up the algorithms' efficiency and accuracy.The sparse transformation matrix estimated from the coarse correspondence will be assigned as the input matrix of b-spline FFD approach.This procedure guarantees the proposed method converge to a better minimum and the registration quality strictly obeys the grid division.That not only improves the efficiency but also the robustness of applying non-rigid registration in removing atmosphere turbulence.In microscopic image application,the conventional microscope system usually rotates the microscope fine focusing knob with the connected electrically controlled motor and then fuses multiple images with diverse focuses into a single one.This method is a highly auxiliary control system related method and increases the system complexity with no common applications.In this thesis,a sensor plane movable camera was designed and a microscopic image feasible automatic fusion algorithm was developed.By adjusting the sensor,the movement accuracy requirement is reduced and more continuous focuses images can be captured which is more favorable for fusing large depth-of-filed image.The developed algorithm applies the clarity factor and pixel selection scheme in image fusion algorithm for the first time,and these applications improved the fusion quality.