Underwater Image Enhancement Based on Double Pyramid Feature Fusion Mode

Abstract

The exponential attenuation of propagated light underwater results in unsatisfactory image features, including color distortion, low contrast and blur. To solve these problems, advantages and disadvantages of image histogram equalization and white balance methods were investigated in this study based on an underwater model, and the method of using the original image to attain the first and second images through histogram equalization and white balance, respectively, was proposed. Laplacian pyramid was introduced to decomposite each image and obtain its layered structure, and the Gaussian pyramid mapping to each weight was formed according to normalized weights. At the end of this study, images were reconstructed by pyramid-based fusion under the guidance of image measurement model and normalized weight mapping model. The experimental results demonstrate that this method integrates the advantages of histogram equalization and white balance methods, providing modified image colors and improved color contrast.