Image Enhancement And Application For Visual Tracking

Image enhancement is a basic topic in the research community of image process-ing. Through processing the image and enhancing the region of interest (ROI), the aimof image enhancement is to get better and more useful images for applications. Differenttechniques of image enhancement are used for different applications. The procedure ofimage enhancement algorithms includes that enhancing the details of images with imagedenoising methods; using image restoration methods to preserve and enhance the struc-ture of image, and getting more image information with given image resolution. Thus theresults of image enhancement can be further used in image-based object recognition andtracking. It can be concluded that image enhancement is a necessary and critical prepro-cessing step for image processing.1. Anewimagedenoisemethodisproposed. Makinguseofredundancyofanimage,we improve the algorithms of non-local means denoising through solving the problemsof huge computation cost. We proposed a multi-scale non-local means method, in whichadaptive multi-scale method is used. Under each selected scale, the input image is dividedintosmallblocks. Thenweremovethenoiseinthegivenpixelusingonlyoneblock. Itcanovercome the low efficiency problem of the non-local means method. Our method is alsobenefit from the local average gradient orientation. Experiments performed in grayscaleandcolorimagesshowthatourmethodisfasterthantheoriginalnon-localmeansmethod.And the results are compared to the output of original and improved non-local meansdenoising methods.2. We propose a new algorithm of document image enhancement to binarize thedocument image. In our approach, the non-local means method is extended and used toremove noises from the input document image in the step of pre-process. Then the pro-posed method binarizes the document image which takes advantage of the quick adaptivethresholding proposed by Pierre D. Wellner. To get more pleasing binarization results, thebinarized document image is post-processed finally. There are three measures in the post-processstep: de-speckle, preservestrokeconnectivityandimprovequalityoftextregions.Experimental results show significant improvement in the binarization of the broken anddegraded document images collected from various sources. 3. A structure-preserving multiscale vessel enhancing diffusion filter is proposed.Enhancement of vessels in medical images is still an unsolved problem. Multiscale ap-proaches were proposed to improve the vessel enhancement effect based on the structuresizeandimageresolution. Vesselenhancingdiffusion(VED)filterisoneofthemultiscaleapproaches, which was based on the scale space theory. VED performs well on enhanc-ing vessel structures but cannot preserve complex structures such as the vessel junctions.In this chapter, a structure-preserving diffusion tensor is defined in the diffusion equa-tion, which brings a structure-preserving vessel enhancing diffusion filter. Through themultiscale framework, the proposed method enhances the vessel structures especially thecomplexstructuresuchasjunctions. Experimentalevaluationperformedonvariousvesseldata sets demonstrated the effectiveness of the proposed method.4. We present a new image quality evaluation method for image denoising. Recentlyproposed methods take different approaches to the problem and yet their denoising per-formances are comparable. However, the performance of the denoising methods has notbeen fully investigated. In this chapter, we first review the existed evaluation measures ofimage denoising methods. Then we give a weight-enhanced evaluation method based onthe analysis of mean square error. We compare our method with other measures on fivestate-of-the-art denoising algorithms. Through the experimental results, the present studywill enable us to understand how well the state-of-the-art denoising algorithms perform.5. We present a new image-based visual tracking method for a robot manipulator totrace a moving target using monocular camera mounted on the end-effector. The intrinsicand extrinsic parameters of the camera are already known. In order to estimate the 3Dposition of the rigid body target, three points (not collinear) from the rigid body wereselected as reference. We assumed that these three points will never be occluded by therigid body itself. Based on this assumption, we employ a new point adaptive UnscentedKalman Filter (UKF) algorithm to track the rigid body and estimate its motion in real-time. Simulation and experimental results are included to illustrate the performance ofthe proposed method on a 3-degree-of-freedom (DOF) robot manipulator.Finally, we conclude the work of this thesis, and give the future work.