Research On Key Technology Of Image Processing In Mobile Augmented Reality And Its Application

With the development of broadband mobile communication technology, as well as the advent of smart mobile phone such as IOS and Android, and mobile applications, the Augmented Reality (AR) that is limited in the labortary before begins to enter the public view, and a large number of terminal location, image recognition technology-based mobile internet application called mobile AR (MAR) applications begin to emerge, and become one of the hotspot in the industry. In 2010, the MAR is ranked in the top ten technology trends that lead the future by the US Times magazine. MAR applys the virtual augmented reality information to the real world, and change the user to observe the world. The traditional WEB page query mode based on text input in the browser will be changed to the next generaton show mode that the hotspot information captured by a variety of sensors is loaded into the real scene in the screen of the mobile, and which will enhance the user experience greatly.In this disseration, the key image processing technology is researched, and then a suit of commercial MAR application system is developed combined with the National Science and Technology Major Project by ZTE Corporation for the practice and exploration of the MAR industralization. In which, the key image processing technology includes image retrieval, object registration and tracing, content rendering and image compression in the MAR application, and then some creative methods are proposed to solve some of the related problems of image searching, image recognization and image rendering in the MAR application.The main subjects researched in this paper are as the following:1. Visual retrievalCondidered the problems of low accuracy and short of bitrate adaptation in visual retrieval of current MAR solution, a robust MAR method for various bitrates with low memory used and easy to distributed deploy is proposed. By selecting feature points based on the statistics and the convergence of multiple efficient affected the performance in image visual retrieval, the linear transformation of simple descriptor and the block-based location point code method, the local feature stream is reduced effectively. Via the Fisher Vector based local descriptor convergence method, the local features are aggregated and the interference factors are removed effectively, and the binary compact descriptor used for distance calculation is generated quickly. As well as, the results are verified with geometric algorithm of DISTRACT and filtered secondly to gain a stable performance with different bitrate streams for the overall visual retrieval system. In addition, the fast linear search method shows its great advantages in the distributed processing and the incremental training to achieve rapid visual retrieval, and provides powerful support for the large-scale deployment of the MAR system.2. Registration & feature trackingA multi-Thread Parallel Computing method for feature matching and tracking is proposed, which optimizes traditional RANSAC algorithm with geometric constraint, and thus effectively solves the lack-of real-time issue of existing AR system based on the unmarked registering & tracking technology. As well as, a quality evaluation function of feature tracking is added to achieve the state switching between threads and the effective screening of the homographic matrix, by which, the system can track fluently irrespective to the object rotation, scaling and occlusion. And finally, a set of AR system is implemented that is available in the mobile terminal with stable and real-time experience.3. Content renderingThe content rendering system of MAR involves key technologies such as script parsing, content loading and presentation. In the paper, to achieve the goal of light weight and high efficiency, some related configuration suggestions are given for choosing the enhanced content in different formats based on massive experimental work. Some more prospective research on the depth of field and shadow rendering algorithm are also studied, which is too complex to use on the mobile intelligent terminal with weak computing capability. For shadow rendering algorithm, the Hashtable is used to promote the efficiency of contour edge finding, besides that, Per-Object algorithm is also used to sharply reduce the number of the triangle slices. For the depth of field rendering algorithm, on the basis of the single-layer algorithm the Gaussian filters are introduced to fuzzy edge to promote the efficiency of the depth of field rendering algorithm. To further improve the operational efficiency of the algorithm, the heterogeneous GPU programming method is used to code implementation, which obtained a more substantial performance improvement.4. Image compressionFirstly, a visual lossless image compression algorithm is realized under the standard JPEG framework, and an effective and complete picture quality evaluation method is introduced for the first time, which dynamically adjusts the quality factor without degrading the user visual perception to obtain a higher compression ratio. In this method, many factors including signal to noise ratio before and after picture compression, edge and texture information are considered, to build an objective picture quality evaluation model. The dynamic adjustment range of quality factors is significantly reduced by using the priori statistical knowledge of quality factors and objective picture quality score; and the system efficiency is effectively improved by using the dichotomy to quickly adjust the quality factor. Secondly, a feature lossless image compression algorithm is implemented. Under the standard JPEG framework, this algorithm adjusts the DCT coefficients to save 42.7% storage compared to the original JPEG algorithm, without losing image feature information. Moreover, the algorithm uses the genetic algorithms to train massive data to obtain the features lossless quantization table, effectively avoiding the problem of big quantization table space, and improving the system efficiency.On the basis of the research, the theoretical results are applied into the engineering practice. After completing the architecture design of the MAR system, the augmented reality service platform and application software are developed, which is a significant practice of mobile Internet value-added services and information services innovation model, and contributes to the promotion of augmented reality service, and promotes the whole industry ecology.