图像融合是图像识别的重要环节，为提升船舶航行图像识别精度，提出了基于视觉传达的大型船舶航行图像融合算法。通过离散滤波获取图像高频部分并盲估计模糊核，依照模糊核，通过总变分正则化法对图像实施非盲去卷积处理，获取去模糊后的图像；针对去模糊处理过程导致图像亮度发生变化的问题，采用伽马校正算法调整图像亮度，提升图像视觉传达效果。针对校正后的图像，对比2幅待融合图像的熵值，获取熵值较大图像的残余分量，通过基于方向滤波的二维局部均值分解法，将残余分量和熵值较小的待融合图像分解成低频子带与高频子带并分别融合，通过逆变换运算得到融合图像。实验结果显示该方法可有效提升大型船舶航行图像的细节清晰度，令图像的视觉传达效果增强，并显著提升图像识别精度。
Image fusion is an important part of image recognition. In order to improve the accuracy of ship navigation image recognition, a fusion algorithm for large ship navigation images based on visual communication is studied. Obtain the high-frequency part of the image through discrete filtering and blindly estimate the fuzzy kernel. According to the fuzzy kernel, perform non blind deconvolution on the image using the total variation regularization method to obtain the deblurred image. To address the issue of changes in image brightness caused by deblurring processing, a gamma correction algorithm is used to adjust the image brightness and improve the visual communication effect of the image. For the corrected image, compare the entropy values of two fused images to obtain the residual components of the image with a higher entropy value. Through a two-dimensional local mean decomposition method based on directional filtering, decompose the residual components and the fused image with a lower entropy value into low-frequency and high-frequency subbands and fuse them separately. The fused image is obtained through inverse transformation operation. The experimental results show that this method can effectively improve the detail clarity of large ship navigation images, enhance the visual communication effect of the images, and significantly improve the accuracy of image recognition.
2023,45(18): 178-181 收稿日期：2023-06-07
DOI：10.3404/j.issn.1672-7649.2023.18.033
分类号：TP391
作者简介：蒋顺生(1982-),男,硕士,工艺美术师,研究方向为艺术设计及数字媒体
参考文献：
[1] 徐志京, 黄海. 基于多重连接特征金字塔的SAR图像船舶目标检测[J]. 激光与光电子学进展, 2021, 58(8): 287-294.
[2] 刘万军, 高健康, 曲海成, 等. 多尺度特征增强的遥感图像船舶目标检测[J]. 自然资源遥感, 2021, 33(3): 97-106.
[3] 贺欣, 周建, 张华. 基于ResNet的红外与低照度可见光图像融合方法[J]. 激光与红外, 2022, 52(11): 1723-1728.
[4] 蒋一纯, 刘云清, 詹伟达, 等. 基于图像退化模型的红外与可见光图像融合方法[J]. 电子与信息学报, 2022, 44(12): 4405-4415.
[5] 罗迪, 王从庆, 周勇军. 一种基于生成对抗网络与注意力机制的可见光和红外图像融合方法[J]. 红外技术, 2021, 43(6): 566-574.
[6] 张浩, 马佳义, 樊凡, 等. 基于特征空间多分类对抗机制的红外与可见光图像融合[J]. 计算机研究与发展, 2023, 60(3): 690-704.
[7] 雷大江, 杜加浩, 张莉萍, 等. 联合多流融合和多尺度学习的卷积神经网络遥感图像融合方法[J]. 电子与信息学报, 2022, 44(1): 237-244.
