| As an important form of information, space motion images play an important role in docking control, docking mechanism design and track interactive control. However, space motion images have different kinds of feature information and one significant feature is the transcale characteristics. It is becoming a new subject in image processing and its research is still at the starting stage. It is necessary to study new methods to analyze and process transcale information. High resolution space image sequences are acquired and the reproduction of high quality space motion could be achieved. From the existing research results at home and abroad, there are three problems in this area, including the change in the attention scale, the change in the frame rate scale and the change in the resolution scale. Based on the background of space motion, this dissertation concentrates the research on the transcale feature and on the image description and interpolation methods for the space motion images. The main contributions and innovations of this dissertation are as follows:(1)Aiming at solving the problem in describing motion images, that is, the image regions attracted by viewers are not depicted, this dissertation proposes a transcale description algorithm for space motion sequences (SITD) using the technique of visual saliency. SITD constructs the attention region calculation framework for space motion image sequences, presents the sustained attention method based on Bayesian probabilistic inference for motion targets of the sequences, and proposes the extraction method based on overlapping rates for transferring frames, so as to capture the characteristic properties of the motion regions easily in the process of describing the space motion images. SITD gains a detection accuracy promotion of10%and8%compared to GCD and ISD, respectively. In order to better describe the motion details, the subtle image motion amplification algorithm (SIMM) is proposed and it could acquire high-quality motion amplification image sequences. SIMM presents a temporal and space processing method based on Eulerian motion magnification, proposes the detection method for space attention regions, and presents the motion detail enhancement scheme, so as to reduce the defects of enlarged results. SIMM could unify the fine scale and coarse scale for space motion image sequences. Experiments have shown that, compared to the EVM, SIMM is more accurate on describing the motion details and it achieves higher objective indices, including meangradient promotion of13%, edge intensity promotion of11%, spatial frequency promotion of27%, power spectrum promotion of4%, and image quality score promotion of14%.(2)Aiming at solving the problem in interpolating the single image, that is, the details of the interpolation results are not clear, this dissertation proposes a novel image interpolation algorithm using saliency detection (SEGPR) for space images, which could obtain the high-resolution interpolation images in accord with the human visual perception. Firstly SEGPR constructs the region mode, which could detect the regions with a high contrast in color, brightness, and edges, and this mode achieves a reasonable regional division. Then it proposes the partition interpolation scheme and it could design a flexible interpolation way according to the region of the interpolated pixel. To highly-ght the motion contents of interpolation images, this dissertation proposes a novel energy-driven interpolation algorithm using Gaussian process regression (EGPR) and it could improve the definition of motion regions and enhance the interpolation quality of motion targets. EGPR constructs the single interpolation computing framework combining edge guiding with statistical theory, proposes the modeling method based on low-resolution image data, and designs the prediction method based on energy computation for unknown pixels, so as to improve the prediction accuracy of interpolated pixels and maintain the detail characteristics effectively. Experiments have shown that, EGPR gains the PSNR promotion of2%,1%and1%in low interpolation factor cases and promotion of5%,2%and2%in high interpolation factor cases compared to BI, GPR and ICBI respectively.(3)Aiming at solving the problems in interpolating image sequences, that is, the motion images showed are not fluent and the details of the interpolation results are blurred, this dissertation proposes a frame interpolation algorithm based on coherency sensitive hashing (CSFI), which could explore temporal and spatial features to depict the characteristics of the edges and textures effectively, so as to obtain the optimal high-resolution interpolation frames. CSFI contains the motion region detection approach based on visual correspondence and the computing of interpolation pixes based on CSH. The former could narrow the prediction range of interpolated frames and the latter could improve prediction accuracy of interpolated pixels contained. Experiments have shown that CSFI gains PSNR promotion of16%,15%,22%,21%and12%and MSSIM promotion of4%,3%,7%,3%and2%, compared to five algorithms including TSS, ARPS, H&S, CNF and CSH. This dissertation proposes a region-guided space motion image sequence interpolation algorithm (RGSI), which could change the resolutions of motion images and obtain the high quality space image sequences. RGSI constructs the regional interpolation strategy and presents the motion detail enhancement method. The former achieves the task of high quality interpolation and the latter strengthens the color contrast of interpolated pixels, which could improve the definition of space motion targets. Experiments have shown that, in the cases of different interpolation factors, RGSI gains edge intensity promotion of24%and17%, and meangradient promotion of23%and15%compared to BI and ICBI, respectively.(4)Aiming at solving the problem in resizing space motion images, that is, the important regions are not changed the scales, this dissertation proposes a new perception driven image resizing algorithm (IPDR), which could improve the definition of motion contents under the guarantee of the global resizing effects. IPDR adopts the partition interpolation method and presents the regional energy protection approach based on seam carving, which could remove the pixels with low energy, so as to improve the definitions of space motion targets in the process of resizing. Experiments have shown that, IPDR gains the meangradient promotion of9%,24%and16%, information entropy promotion of3%,17%and7%, and power spectrum promotion of12%,11%and13%, compared to IR, IC and ICBI, respectively.(5)This dissertation proposes a transcale interpolation model for space motion images (MTIM) based on previous studies including SIMM, SITD, SEGPR, CSFI, RGSI and IPDR. The MTIM model contains the following algorithms:the motion target course description algorithm, the key frames transcale exhibition and the movement display exhibition algorithm. And it realizes the course description of space motion targets under the different scales, and the display of key frames under the different time scales. Experiments have shown that, MTIM could obtain good transcale interpolation results and display high quality transcale exhibition for space motion image sequences. |
PDF Full Text Request