An infrared night vision image enhancement algorithm based on cross-level feature fusion

Infrared night vision images are caused by color overflow and coloring discontinuity due to insufficient light at night, resulting in larger halo area and lower PSNR value after enhancement by single feature fusion method. For this reason, an infrared night vision image enhancement algorithm based on cross-level feature fusion is proposed. This method is used to denoise infrared night vision images, based on smooth wavelet decomposition. By labeling image edges and noise, and utilizing neighborhood based wavelet coefficient shrinkage algorithm, the noise interference in the image is effectively reduced; preliminary enhancement was performed on the denoised image, using Retinex algorithm combined with bilateral filtering method to estimate illuminance, and Sigmoid function was used to enhance the reflection area, improving the overall visual effect of the image. Based on the principle of cross-level feature adaptive fusion, a cross-level feature fusion network is constructed to further enhance the feature information of the infrared night vision image through the steps of multi-level feature extraction, feature reconstruction and adaptive cross-level feature fusion, and the output of the model is optimized by using the joint loss function, which realizes the high-quality enhancement of the infrared night vision image. The experimental results show that when the method is utilized for infrared night vision image enhancement, the PSNR is higher than 30dB, the SSIM is higher than 0.73, and the enhancement effect is good and the performance is high.