Multispectral Image Enhancement Based on the Dark Channel Prior and Bilateral Fractional Differential Model

Weijie Chen,Jie Yang,Nikola K Kasabov,Zhenhong Jia

doi:10.3390/rs14010233

Weijie Chen, Jie Yang + Show 2 more

Open Access

https://doi.org/10.3390/rs14010233

Copy DOI

Abstract

Compared with single-band remote sensing images, multispectral images can obtain information on the same target in different bands. By combining the characteristics of each band, we can obtain clearer enhanced images; therefore, we propose a multispectral image enhancement method based on the improved dark channel prior (IDCP) and bilateral fractional differential (BFD) model to make full use of the multiband information. First, the original multispectral image is inverted to meet the prior conditions of dark channel theory. Second, according to the characteristics of multiple bands, the dark channel algorithm is improved. The RGB channels are extended to multiple channels, and the spatial domain fractional differential mask is used to optimize the transmittance estimation to make it more consistent with the dark channel hypothesis. Then, we propose a bilateral fractional differentiation algorithm that enhances the edge details of an image through the fractional differential in the spatial domain and intensity domain. Finally, we implement the inversion operation to obtain the final enhanced image. We apply the proposed IDCP_BFD method to a multispectral dataset and conduct sufficient experiments. The experimental results show the superiority of the proposed method over relative comparison methods.

Highlights

Multispectral remote sensing images have been widely used in agriculture, forestry, mineral exploration, military, and many other fields, resulting in huge social and economic benefits [1]; due to the limitations of sensors and atmospheric scattering, the visual effect and spatial resolution of multispectral remote sensing images cannot fully meet the demands of people; image enhancement processing is usually used before image analysis and interpretation to highlight useful information and expand the differences between different features [2,3,4]
By referring to the relevant works of multispectral remote sensing image enhancement [30,31,32,33,34,35,36,37,38], five well-known evaluation indexes, including the contrast, image intensity, information entropy, average gradient, and execution time are used to evaluate the performance of different methods
A multispectral image enhancement method based on dark channel prior technology and the fractional differential algorithm is proposed

Summary

Introduction

Multispectral remote sensing images have been widely used in agriculture, forestry, mineral exploration, military, and many other fields, resulting in huge social and economic benefits [1]; due to the limitations of sensors and atmospheric scattering, the visual effect and spatial resolution of multispectral remote sensing images cannot fully meet the demands of people; image enhancement processing is usually used before image analysis and interpretation to highlight useful information and expand the differences between different features [2,3,4]. Multispectral remote sensing images are generated by collecting several bands of the same region in different spectral sampling intervals [5]. The generated data include information from multiple channels. Single-channel image enhancement methods mainly include spatial domain algorithms and frequency domain algorithms [6,7,8]. Common spatial domain algorithms include histogram matching [9,10,11], Retinex algorithms [12,13,14,15], morphological methods [16], differential filtering algorithms, dark channel prior algorithms, and deep learning algorithms. In integer-order differential algorithms, several integer-order operators, including first-order operators, such as the Sobel operator and Prewitt operator, and second-order operators, such as the Laplacian operator [17], have been proposed to Remote Sens.

Methods

Results

Discussion

Conclusion