Research On Image Quality Improvement Algorithms For Infrared Imaging

Infrared imaging systems can acquire the passive image of the thermal target scene and have the characteristics of good concealment,non-contact,and full-time work so that they are widely applied in military and civilian fields.However,due to the influence of innate factors including the thermal imaging mechanism,the atmospheric transmission and the fabrication process of infrared detectors,the degradation of infrared images with low signal-to-noise ratio,bad contrast and blurred edges and details is a common problem for infrared imaging systems.As a result,infrared image quality improvement technology including various pretreatments is developed according to the characteristics of infrared imaging systems and infrared images.Excellent algorithms for improving the quality of infrared images would enable the resulting images to have better visual effects and more information.Therefore,infrared image quality improvement technology is one of the key technologies in modern infrared imaging systems,and developing the relevant algorithms is of great theoretical and practical significance to performance improvements of infrared imaging systems.This dissertation focuses on two significant aspects,nonuniformity correction and image enhancement,in the study of infrared image quality improvement.Specifically,some key issues of infrared image preprocessing are deeply studied,including stripe nonuniformity correction,dynamic range compression and detail enhancement,and blur removal and contrast enhancement.And the related algorithms are proposed.The main contributions of this dissertation are summarized as follows.1.A spectral-spatial filtering algorithm is proposed for stripe nonuniformity correction.After analyzing the spatial and spectral characteristics of stripe nonuniformity noise,a two-stage filtering strategy is designed.In the first stage,with a rectangular notch filter,the original infrared image is processed by spectral filtering.As a result,it is decomposed into a clean structural layer and a striped background layer.In the second stage,stripe nonuniformity noise is filtered out from the background layer by an iterative operation of 1-D row smooth filtering.The final corrected image is obtained by merging the two layers resulting from the two filtering stages.The experimental results demonstrate that the proposed algorithm is efficient and effective,and it produces better correction results,compared with other conventional and advanced stripe nonuniformity correction algorithms.2.In order to correct the stripe nonuniformity while preserving more useful edge structure information for infrared images,an improved algorithm based on frequency domain anomaly detection and spectral fusion is proposed.By analyzing the spectral statistics of infrared images,it is found that the average power spectrum is modeled as a generalized Laplacian distribution.First,after the ideal image spectrum is estimated with the prior model,the abnormal frequencies caused by stripe nonuniformity noise are designedly detected according to the spectral distribution of the noise.Then,the anomalies are corrected by fusing the spectrums of the original image and a stripe-free smooth reference image with edge preservation.Finally,the correction result is reconstructed with the fused Fourier spectrum.The experimental results demonstrate that the proposed algorithm has better robustness and edge-preserving ability than the algorithm presented in Section 1,and it consistently produces better correction results and can cope with many complex situations on stripe noise,compared with other conventional and recently deep-learning-based stripe nonuniformity correction algorithms.3.A detail enhancement algorithm based on multiscale guided filtering is proposed for 14 bit or 16 bit high dynamic range infrared images.After analyzing the principle of guided filter,the architecture of multiscale guided filtering is designed to extract detail features of image.First,the original image with high dynamic range is decomposed by multiscale guided filtering.Then,different layers are processed individually through dynamic range adjustment and optimization according to their characteristics,thus enhancing image contrast and details.Finally,the enhanced result is reconstructed from layers.The experimental results demonstrate that the proposed algorithm has excellent performance of detail extraction and enhancement,and it produces high-quality results with better visual effects than other conventional dynamic range compression algorithms and representative detail enhancement algorithms.In addition,the proposed algorithm is directly applicable for 8bit infrared images.4.A contrast enhancement algorithm based on atmospheric scattering model is proposed for infrared images.After analyzing atmospheric scattering model which physically describes the degradation of hazy images,the model is applied to remove the fogging effect of infrared images.First,the ambient light and atmospheric light as the model variables are estimated separately by a weighted least squares model and a block statistics strategy.Then,two postprocessing enhancement steps are designed to improve the global contrast and feature visibility of the haze-free image.The experimental results demonstrate that the proposed algorithm has outstanding abilities to remove the fogging effect and enhance the contrast of image,and it produces high-quality results with better visual effects than other conventional and advanced enhancement algorithms.