A Variational Model for Sea Image Enhancement

Mingzhu Song,Guang Jin,Hongsong Qu,Guixiang Zhang,Shuping Tao

doi:10.3390/rs10081313

Abstract

The purpose of sea image enhancement is to enhance the information of the waves, whose contrast is generally weak. Enhancement effect is often affected by impulse-type noise and non-uniform illumination. In this paper, we propose a variational model for sea image enhancement using a solar halo model and a Retinex model. This paper mainly makes the following three contributions: 1. Establishing a Retinex model with noise suppression ability in sea images; 2. Establishing a solar-scattering halo model through sea image bitplane analysis; 3. Proposing a variational enhancement model combining the Retinex and halo models. The experimental results show that our method has a significant enhancement effect on sea surface images in different illumination environments compared with typical methods.

Highlights

Sea surface image enhancement is an important step in image preprocessing of marine remote sensing
Histogram equalization-based enhancement methods are based on the idea of stretching the dynamic range of an observed image, from inchoate global full gray-level histogram stretching methods [1,2] to the contrast-limited adaptive histogram equalization (CLAHE) method [3] and fast hue and range-preserving histogram method [4] proposed in recent years
We use the neighborhood mean value Ia of the central pixel and the neighborhood mean value Ib of the edge pixels in the image as the two extreme values of the Gaussian function, Ia is defined as the mean value of the elliptical center pixel and its surrounding 17 × 17 pixels, and Ib is defined as the mean value of Comparing Figure 4 with Figures 2 and 3, our model shows consistency with the actual image

Summary

Introduction

Sea surface image enhancement is an important step in image preprocessing of marine remote sensing. We use the neighborhood mean value Ia of the central pixel and the neighborhood mean value Ib of the edge pixels in the image as the two extreme values of the Gaussian function, Ia is defined as the mean value of the elliptical center pixel and its surrounding 17 × 17 pixels (if the region contains abnormal non-uniform noise, the abnormal point is removed), and Ib is defined as the mean value of the edge pixel and its surrounding 17 × 17 pixels (if several edge pixel regions can be found, the mean of all regions is used).and we can obtain the final solar-scattering halo fitting model: S(x, y) = H(x, y) · (Ia − Ib). When a sea image contains only a part of the elliptical halo, we intercept the corresponding position area in the model above and obtain the required halo components in the subsequent processing

Final Variational Model

Experiment Settings

Sea Image Enhancement

Findings

Further Correction

Conclusions