Abstract
This paper proposes an efficient method to modify gradient and intensity histograms (GIH) for contrast enhancement, which plays an important role in remote sensing image processing and information extraction. First, a self-adaptive algorithm is used to flatten the shape of GIH of input image according to standard deviation of GIH. Then, the standard lookup table-based histogram equalization procedure is applied to get well enhanced image. Experimental results, using various remote sensing images, show that the proposed method generates enhanced images with more information and higher visual quality, compared with several conventional methods.
Introduction: The Challenge of Remote Sensing Contrast
Remote sensing images—captured from satellites or high-altitude drones—are essential for applications ranging from agricultural surveillance to military reconnaissance. However, these images are often plagued by low contrast due to adverse weather conditions, insufficient sunlight, or sensor limitations.
Standard enhancement techniques, most notably Histogram Equalization (HE), often fall short. HE works by redistributing pixel intensities to cover a wider range, but it has a significant flaw: it over-enhances regions with high pixel counts. This leads to visual artifacts and a loss of detail in extremely bright or dark areas, often resulting in a "blown-out" appearance. While newer methods like Gradient and Intensity Histogram Equalization (GIHE) incorporate structural data, they still struggle with "high peaks" in the histogram that cause over-stretching. This research introduces an automated "flattening" technique to achieve natural, detail-rich results without manual tuning.
Proposed Methodology: GIH and Self-Adaptive Flattening
The authors propose a two-step self-adaptive approach designed to prioritize structural integrity over simple pixel statistics.
1. Gradient and Intensity Histogram (GIH) Construction
Instead of merely counting pixel frequencies at each brightness level, the GIH weights each pixel by its gradient—the rate of change in brightness. This ensures that structural edges and textures are prioritized over flat, uniform areas.
2. Self-Adaptive Flattening
To prevent the over-enhancement typical of HE, the authors apply a power-law transformation to the histogram: $GIH_f(k) = GIH(k)^p$. The innovation lies in the parameter $p$.
- If $p=1$, the method functions as standard GIHE.
- If $p=0$, it results in a simple linear stretch.
The algorithm automatically calculates the optimal $p$ based on the standard deviation of the GIH. This allows the system to "sense" the required degree of flattening for any specific image, finding the "sweet spot" between under-enhancement and harsh, artificial-looking results.
Experimental Results and Comparison
The proposed method was evaluated against industry standards, including Global Histogram Equalization (GHE), Adaptive Gamma Correction with Weighting Distribution (AGCWD), and Bi-Histogram Equalization with Gradient and Intensity (BP-GIHE).
Qualitative Analysis
In various test cases, such as images of Mars, islands, and hilly terrain, the proposed method consistently outperformed traditional techniques. For instance, in "Island" imagery, it preserved the delicate textures of sea spray (spindrift) that other methods turned into featureless white blobs.
Quantitative Analysis
Performance was measured using Discrete Entropy (E). While raw images often have high entropy due to their uncompressed state, they are frequently unreadable. The goal is to maximize entropy while achieving clear visual contrast. The proposed method consistently achieved the second-highest entropy among all tested enhancement methods, striking a balance between information retention and visual clarity. For the "Island" image, the proposed method scored 5.9955, significantly outperforming traditional GHE (5.2041) and AGCWD (5.7813).
Conclusion and Future Directions
By prioritizing structural information (gradients) and automating the histogram flattening process, this method provides a robust solution for remote sensing image enhancement. It avoids common pitfalls like over-enhancement and artifact creation. Future work may explore the computational efficiency of gradient calculation for real-time applications and extend the algorithm to handle multi-spectral or infrared data.
Summary updated