Curvature filter and gradient transform based image enhancement algorithm can effectively suppress noises and enhance image edges. However, it is very hard to be carried out in real time due to the large computing load. To address this problem, a GPU based parallel implementation is proposed in this paper. First, aiming at the characteristics of the algorithm, a numerical implementation method based on central-difference is proposed. Then a domain decomposition scheme is utilized in parallel Gaussian curvature filter to remove the dependence of neighboring pixels and guarantee convergence. Finally, we make the multiprocessor wrap occupancy reach 100% by optimizing the thread grid and register usage. Experimental results demonstrate that our parallel method runs 200-300 times faster than CPU serial method with real time processing of 4096×4096 resolution image, which indicates a great potential for application.
Hyperspectral image is a three-dimensional data cube which describes spatial information and spectral information of the scene. The anomaly detection technique can detect the targets which have difference between the image and the background without priori information. Kernel independent component analysis(KICA) is a method of mapping hyperspectral data into the kernel space for feature extraction. In this paper, the hyperspectral image is subjected to abnormal information detection based on KICA. First, we calculate the kernel matrix K in order to map the data to high-dimensional space for whitening and dimension reduction processing. Then we utilize the FastICA algorithm to extract the core independent component (KIC). Finally, the extracted independent components with the most abnormal information are analyzed by RX operator, kernel RX operator and abundance quantization method. Comparing with the simulation result and the detected result by RX method, the representation shows the algorithm based on KICA has better detection performance.
In this paper, a curvature filter and PDE based non-uniformity correction algorithm is proposed, the key point of this algorithm is the way to estimate FPN. We use anisotropic diffusion to smooth noise and Gaussian curvature filter to extract the details of original image. Then combine these two parts together by guided image filter and subtract the result from original image to get the crude approximation of FPN. After that, a Temporal Low Pass Filter (TLPF) is utilized to filter out random noise and get the accurate FPN. Finally, subtract the FPN from original image to achieve non-uniformity correction. The performance of this algorithm is tested with two infrared image sequences, and the experimental results show that the proposed method achieves a better non-uniformity correction performance.
The target is moving and changing in infrared image sequences captured from the airborne platform infrared imaging system. To adaptively track the infrared target which changes from small target to surface target, an algorithm based on Second-Order Differential (SOD) and improved Template Matching (TM) tracking algorithm was proposed. The SOD filter makes full use of the brightness of the infrared dim and small target, the gradient and distance information of neighborhood pixels used for spatial domain filter. The TM makes full use of infrared brightness, ambient background and dimension information to complete the tracking. The experimental results show that the proposed algorithm can convert adaptively with infrared target’s size changing information, so tracking stability of infrared target under the ground clutter background is achieved. The tracking accuracy and tracking speed are also better than traditional algorithms. The proposed algorithm can be well applied to airborne platform warning on the ground.
In this paper, a new temporal high-pass filter nonuniformity correction algorithm based on guided filter is proposed, which address the ghosting artifacts and preserve image details of original image. In this algorithm, the original input image is separated into two parts, which are the high spatial-frequency part that contains most of the nonuniformity and the low spatial-frequency part with well preserved details. Then the fixed pattern noise is estimated from the high spatial-frequency part and subtracted from the original image, which achieves the nonuniformity correction. The performance of this presented algorithm is tested with two infrared image sequences, and the experimental results show that the proposed algorithm can significantly reduce the ghosting artifacts and achieve a better nonuniformity correction performance.