No-reference Quality Evaluation For Infrared Image And Its Application

The quality evaluation of infrared images has important theoretical significance and engineering value. The quality of infrared images is an important metric determining performance of infrared imaging systems. However, the effective quality evaluation of the inputted infrared images is a prerequisite for the performance evaluation of infrared image processing algorithms. The quality evaluation of infrared images is special in that there is no reference image for the most of infrared images, which is within the scope of no reference image quality evaluation. The current methods to evaluate the quality of infrared images have not universality, are independent of each other, and don’t constitute a recognized and complete evaluation system.The main content of this dissertation is to explore and establish an efficient quality evaluation system of infrared images without reference image and associated evaluation methods. In addition, image segmentation is a basis of numerous algorithms in image processing(such as target detection, feature analysis, target tracking, and precise positioning). The significance of image segmentation promotes us to make some associated studies on the applications of the quality evaluation of infrared images in image segmentation. A novel quality evaluation system of infrared images is proposed. Compared with the current evaluation methods, the novel system has the following advantages:(1) there is a good complementary among the evaluation methods, constituting a complete quality evaluation system of infrared images;(2) in regard to different applications of infrared images(man in loop, and automatic target recognition), methods to evaluate the quality of infrared images are established in accordance with different technical lines(based on human vision mechanism, and associated with image processing algorithms) respectively; and(3) the evaluation methods fully consider the characteristic that there is no reference image during the quality evaluation. Innovations of this dissertation are summarized as follows:1. In regard to quantifying background clutter in infrared images, we propose a method to quantify background clutter in infrared images based on human vision features by comparing the salient low-level features which are the phase congruencies(PCs), gradient magnitudes(GMs) and the like, between respective background regions in images and target regions. The experiments show that the proposed method of quantifying clutter can measure accurately the strength of background clutters in infrared images, the measurement results fit human visual perception well, and the quantitative results fit actual situation better than the traditional quantitative metrics of clutters.2. By analyzing the factors in infrared images which interfere with performances of target detection, a method to evaluate the quality of infrared images based on target detection performances is proposed. In order to quantitatively describe this method, the quantitative description of interference degree in background using interference degree of global background(IDGB) and similarity degree of local background(SDLB), is defined and designed. The experimental results show that the evaluation results of the proposed metrics fit actual situation better than that of the traditional metrics.3. In relation to infrared image sequences, the factors in sequence which interfere with target-tracking are analyzed, and a conclusion that intra-frame image quality and inter-frame image quality in infrared sequence images will commonly affect the effect of target tracking is made. Based on the conclusion, three novel evaluation metrics of infrared image sequence are proposed, namely, intra-frame degree of target being shielded(IFDTS), inter-frame change degree of target gray information(IFCDTGI), and inter-frame change degree of target motion information(IFCDTM), and specific definitions and detailed calculation methods are given. The experiments show that the proposed metrics which are suitable for using in the quality evaluation of the infrared image sequence, correlate well with the performance of tracking algorithms, and present monotonic relation with probability, is an effective method of evaluating quality of the infrared image sequences. This work compensates for the shortages on the quality evaluation in the field of performance evaluation of infrared images processing algorithms.4. In regard to the application of the quality evaluation method of infrared images in image segmentation, factors in infrared images disturbing the performance of segmentation algorithm are analyzed, activity degree of local area(ADLA) is proposed to quantitatively describe the above described factors, and then the related studies are carried out in the two following aspects:(1) The key parameters of Stewart PCNN are determined self-adaptively using local activity of images, thereby solving the drawback that the key parameters of models need to be tested artificially and be determined repeatedly.(2) The prior performance evaluation of image segmentation algorithms often ignore the influence of image quality, and doesn’t make much sense for the performance evaluation of algorithms, and the comparison and selection among different algorithms. According to the above described drawbacks, we propose activity degree of image(ADI) based on local activity of images, and then construct a method to evaluate the performance of segmentation algorithms based on image quality, namely weighted excellent degree(WED), combined with excellent degree(ED), using the metric ADI.The work of this dissertation builds a nice basis for further developing the work of the quality evaluation of infrared images without reference image.