A System Noise Removal Method for Series Images from Spaceborne Area Array Camera

Objectives: Due to the difficulty of the on-orbit radiometric calibration of area-array camera, a large amount of system noise will appear in the area-array imaging remote sensing images when the relative radiometric calibration coefficients are not accurate enough or the satellite state changes. Most of the existing denoising methods are based on the spatial domain correlation or transform domain correlation of images, and the denoising images will have the problem of losing edge details and texture characteristics. We analyzes the source of the system noise and proposes a method to remove system noise using series images based on the correlation of system noise in the time axis of series images. Methods: First, the original image and the Gaussian filtered image are firstly combined in a ratio operation to eliminate the radiation characteristics to obtain multiple texture images. Second, the circular local binary pattern (LBP) operator and Grubbs criterion are combined to eliminate the gradient characteristics to obtain multiple noise images. Third, the correction coefficients are obtained by the superposition of multiple noise images, and further used to filter out the system noise. Results: The simulated and real images are used as the experimental data. Our proposed method was compared with non-local mean(NLM), discrete cosine transform(DCT), wavelet thresholding denoising, weighted nuclear norm minimization(WNNM), and block matching 3-D filtering(BM3D). Our method obtains the highest peak signal-to-noise ratio(PSNR) and structural similarity index(SSIM) with different noise levels in simulation experiments. For real data experiment, the laboratory coefficient correction, DCT and wavelet thresholding can not remove the noise well. BM3D and WNNM can efficiently remove noise and obtain high signal-to-noise ratio(SNR), but the edge and texture details are lost seriously. Our method removes the noise very well, and it can maintain the edge details and texture information. Conclusions: The results show that the method in this paper can effectively maintain image edge details and texture information while removing image noise.