Fractional-Order Integration Based Fusion Model for Piecewise Gamma Correction Along With Textural Improvement for Satellite Images

• Published in: IEEE access 2019, Vol.7, p.37192-37210
• Main Authors: Singh, Himanshu, Kumar, Anil, Balyan, L. K., Lee, Heung-No
• Format: Article
• Language: English
• Subjects: Adaptive filters; Algorithms; Calculus; Filtering; Fitness; Fractional-order (FO) masking filter; fractional-order integration (FOI); gray-level co-occurrence matrix (GLCM); Histograms; Homogeneity; Image contrast; Image edge detection; Image enhancement; Image filters; Image quality; Image restoration; Machine learning; opposition-based learning (OBL); optimal mask designing; Optimization; Performance evaluation; piecewise-gamma correction (PGC); quality enhancement; Riemann–Liouville (RL) definition; Satellite imagery; Sharpness; sine cosine algorithm (SCA); Texture; Trigonometric functions; two-dimensional (2-D) adaptive filtering
• ISSN: 2169-3536; 2169-3536
• DOI: 10.1109/ACCESS.2019.2901292

Fractional-order integration (FOI) and its beauty of optimally ordered adaptive filtering for image quality enhancement are latently too valuable to be casually dismissed. With this motivation, a new Riemann-Liouville fractional-order calculus-based spatial-masking methodology is proposed in this paper in association with counterbalanced piecewise gamma correction (PGC). A generalized FOI-based mask is also suggested. This mask is negatively augmented with the original image for harvesting texture-based benefits. PGC is just employed through a constructive association of both kinds of reciprocally dual gamma values (\gamma _{1}= \gamma and \gamma _{2}=1/\gamma, \forall \gamma > 1 ), which leads to optimally desired enhancement when applied in a weighted counter-correction manner. Efficiently improved and recently proposed opposition-based learning inspired sine-cosine algorithm is employed in this paper, along with a newly framed fitness function. This fitness function is devised in a novel manner by taking care of textural as well as non-textural details of the images. In this paper, especially for dark images, 130% increment is achieved over the input contrast along with the simultaneous 147% increment in the discrete entropy level and 22.8% increment in the sharpness content. Also, brightness and colorfulness are reported with 130% and 196.4% increased with respect to the input indices, respectively. In addition, the textural improvement is advocated in terms of desired comparative reduction of gray-level co-occurrence matrix-based metrics, namely, correlation, energy, and homogeneity, which are suppressed by 25.6%, 72.5%, and 21.8%, respectively. This performance evaluation underlines the excellence and robustness for imparting proper texture as well as edge preserved (or efficiently restored) image quality improvement.