While deep learning methods show promising visual results,current end-to-end networks often lack tailored architectures to address common issues like color distortion and texture blurriness. To im-prove their effectiveness,we propose an underwater image enhancement network that utilizes joint texture perception and color histogram features. The network comprises a texture-aware module,a color histo-gram extraction module,and a color-texture fusion enhancement module. The texture-aware network in-corporates a deformable transformer module,leveraging spatially aware deformable convolution to en-hance multi-head attention and extract texture features. The color histogram extraction module harnesses histograms from real underwater images to compute the loss function. Subsequently,the color-texture fu-sion module merges the color and texture features,which are then processed by the enhancement network to produce the final results. This approach effectively preserves texture structures,corrects color distor-tions,and maintains information consistency. Extensive experiments demonstrate that our method surpass-es existing underwater image enhancement algorithms,achieving a 10％ increase in the UIQM metric and reducing processing time to just 0.051 s per image. Our model successfully meets the demands of underwa-ter visual enhancement tasks.
维普
万方数据