Ant Cuckoo Search Optimization‐based deep learning classifier for image enhancement in spinal cord images. Issue 4 (27th May 2021)

• Record Type: Journal Article
• Authors: Saminathan, Priya; Samuel, Letitia
• Is Part Of: International journal of imaging systems and technology. Volume 31:Issue 4(2021)
• Page Start: 2267
• Page End: 2282
• Publication Date: 2021-05-27
• Subjects: Ant Lion Optimization -- circular‐based searching method -- Cuckoo Search Optimization -- image enhancement -- Image Enhancement Conditional Generative Adversarial Network; Imaging systems -- Periodicals; Image processing -- Periodicals; 621.367
• DOI: 10.1002/ima.22597
• Languages: English
• ISSNs: 0899-9457

Abstract

Abstract: Image enhancement has been paid great attention in several areas, and various image enhancement approaches are introduced by the researchers to enhance the images clearer from the degraded images. Accordingly, this paper presents the proposed ACSO‐based IE‐CGAN for image enhancement. Initially, an effective noisy pixel identification based on image enhancement is developed, which employs circular‐based searching for improving the input image from several noises, like salt and pepper noise, speckle (impulse) noise, Gaussian noise, and random noise. Once the noisy pixels are identified, the pixel enhancement is done using the circular‐based search. The noise removal is done using a statistical model, where the best threshold value is determining using the proposed Ant Cuckoo Search Optimization (ACSO) algorithm. The ACSO algorithm is newly designed by integrating Ant Lion Optimization (ALO) and the Cuckoo Search Optimization (CSO) algorithm. After that, the contrast enhancement is carried out using Image Enhancement Conditional Generative Adversarial Network (IE‐CGAN), which is trained by the proposed ASCO algorithm. The experimentation is carried out using an osteoporotic vertebral fracture dataset, and the performance of image enhancement using ACSO‐based IE‐CGAN is evaluated based on Peak Signal to Noise ratio (PSNR), Structural Similarity Index (SSIM), and Second Derivative like Measure of Enhancement (SDME). The developed method achieves the maximal PSNR of 33.50 dB with salt and pepper noise, maximal SDME 41.38 dB with random noise, and maximal SSIM of 0.967 with salt and pepper noise.