Identification of source of image degradation using a new hybrid dense F-LSTM network and image restoration

Image degradation has been garnering huge interest from researchers in the field of image processing and computer vision as images may contain several degradations that lead to different types of blurring, noise, and distortions. Hence, a new hybrid optimized network named DenseNet Fused Long Short Term Memory (Dense F-LSTM) is proposed for the identification of the source of image degradation. Initially, the input image with noise and blur is taken from the database and is subjected to noisy pixel identification, and identification of image source degradation, where the noise pixel identification is performed by Deep Convolutional Neural Network (DCNN), and the image source degradation is employed by proposed Dense F-LSTM. The identified noisy pixel is fed to the Type 2 Fuzzy and Cuckoo Search (T2FCS) filter for noisy correction for obtaining the denoised images. Once the image source is identified, the identified source of image degradation and the denoised image is used for performing deblurring. Here, deblurring is done by kernel estimation, which is optimally selected employing Jaya Walrus Optimization (JWaO). The JWaO is formed by the integration of the Jaya Algorithm (JA) and Walrus Optimization Algorithm (WaOA). Both the denoised image and deblurred image are fused to gain the final output image. The devised Dense F-LSTM method provided superior identification performance and achieved a Second-Derivative-like Measure of Enhancement (SDME) is 59.956, Peak Signal Noise Ratio (PSNR) is 49.924 dB and Structural Similarity Index Measure (SSIM) is 0.979, and Mean Squared Error (MSE) of 0.2.