To address the issue of stripe noise in the overlap regions of multispectral images captured by focal plane modules (FPMs) with charge-coupled devices (CCDs), we propose a method to reconstruct the radiometric model in these areas by synchronizing the correction of overlapping pixels with the normalization of radiometric data. After correcting the same pixels within the overlap region, the radiometric model is rebuilt by summing the detected energy and applying a moment-matching technique to ensure smooth transitions along the region’s edges. The method proposed in this paper reduces image saturation in the overlap regions post-correction and effectively addresses stripe noise. In addition, it optimizes the stripe coefficient to be more responsive in affected regions without influencing non-affected regions, enabling a quantifiable analysis of stripe energy. The experimental results show significant reductions in stripe noise or even its complete elimination, a decline in stripe coefficients, and a notable improvement in image radiometric consistency.

When capturing the same object, incident light is split, resulting in the overlap area detector receiving less incident energy and a lower signal-to-noise ratio (SNR) than the non-overlap area detector. This leads to inconsistent responses and greater susceptibility to stripe noise in the overlap area. In this paper, we propose a relative radiometric correction method that reconstructs energy in these regions by combining the radiometric values from adjacent CCDs. By superimposing the digital number (DN) of pixels with the same name from adjacent CCDs, this approach restores the average DN of the overlap area detector to align with that of non-overlap areas, thus enhancing SNR and response consistency while reducing stripe noise. The basic principle is illustrated in Fig. 1.

In the gobi in Dunhuang and Qinghai Lake regions, generalized noise in overlap area images after correction is reduced to 4.5‰ and 4.1‰, respectively, well below the 3% threshold, demonstrating the efficacy of this energy reconstruction method in removing strip noise while maintaining radiometric accuracy. The radiometric uniformity in the overlap areas across the regions is compared, demonstrating that the proposed method effectively eliminates stripe noise, resulting in a significant improvement in image quality (Fig. 11). A comparison between Figs. 12 and 13 reveals that the stripes in images processed by the proposed method have disappeared, significantly enhancing image quality. Detailed comparisons of the two methods, based on the average DN of pixels at fringe positions, are provided in Fig. 14. As shown, the histogram statistics method results in pronounced spikes in the average DN when detecting streaks. In contrast, the proposed method yields a relatively smooth average DN in fringe areas, eliminating visible fringes in the image. Using the traditional histogram-based matching method, the peak values of fringe coefficients in overlap areas for city, desert, and snow mountain images are approximately 50, 15, and 70, respectively. However, the proposed method reduces these peaks to around 10, 3, and 15, respectively. Fig. 15 further illustrates that the peak position of the fringe coefficient corresponds to the location of the fringe. As shown, the fringe coefficient in overlap areas is notably lower with the proposed method, eliminating visible fringes and resulting in superior image quality.

In this paper, we reconstruct the radiometric model in overlap areas by employing same-pixel matching and summation of detection energies, followed by a relative radiometric correction. Transitional color leveling is then applied to overlap edges using the moment-matching method, which achieves effective relative radiometric correction. The method addresses stripe noise in multispectral image overlap regions, reduces saturation, and ensures uniformity post-correction. Compared to traditional histogram-based matching methods, this method significantly minimizes or eliminates stripe noise in overlap areas. In tests with city, desert, and snow mountain images, peak strip coefficients drop from 50, 15, and 70 to 10, 3, and 15, respectively, enhancing radiometric consistency and image quality. This method shows great potential for broad applications in optical image processing.