摘要:
原始高空间分辨率海岛海岸带遥感影像往往存在影像灰暗、偏色、地物信息较难辨识的现象。为及时获取清晰、信息丰富、反差适中、亮度均匀的海岛礁遥感影像,满足日益强烈的海岛海岸带地理信息保障需求,针对海岛海岸带高空间分辨率遥感影像,该文提出一种深度学习结合改进直方图匹配的智能化调色方法。首先,进行数据重采样与自适应分块获取抽稀影像; 其次,应用MBLLEN网络对抽稀影像进行真彩色增强; 最后,采用改进直方图匹配的方法对原始影像进行色彩映射,最终得到符合人眼视觉、色彩一致、细节丰富的遥感影像。采用主客观相结合的方式综合评价调色效果,结果表明: 相较于Retinex,HE和MASK等常用调色方法,该文算法结果更符合人眼视觉、色彩一致、细节丰富,可有效改善海岛海岸带高空间分辨率遥感影像视觉效果,较好地保留原始地物的细节信息,大幅提升调色效率。
Abstract:
The original high-spatial-resolution remote sensing images of coastal zones of islands often exhibit a gray tone, color cast, and indistinguishable surface feature information. In response to the increasing demand for geographic information security of coastal zones of islands, this study aims to obtain timely clear remote sensing images with rich information, moderate contrast, and uniform brightness for island reefs. Hence, it proposed an intelligent color enhancement method by combining deep learning with improved histogram matching for high-spatial-resolution remote sensing images of coastal zones of islands. First, data resampling and adaptive chunking were performed to obtain thinned images. Then, the MBLLEN network was applied to enhance the thinned images with true color. Finally, an improved histogram matching method was employed for color mapping of original images, obtaining remote sensing images with consistent colors and rich details conforming to human vision. The color-matching effects of these obtained remote sensing images were evaluated using both subjective and objective methods. The results show that compared to other commonly used color-matching methods like Retinex, HE, and MASK, the method proposed in this study yielded more satisfactory results characterized by consistent colors and rich details conforming to human vision. Therefore, the proposed method can effectively improve the visual effects of high-spatial-resolution remote sensing images of coastal zones of islands, effectively retain the details of original surface features, and significantly enhance color-matching efficiency.
下载: