Single image super-resolution with self-organization neural networks and image laplace gradient operator

At present, artificial neural networks have received wide applications in the field of image processing and image resolution because of their fast algorithm implementation and their high accuracy. Learning-based super-resolution methods used stochastic computation in their algorithms, leading to a manual and experimental adjustment of the regularization parameter to the solving imaging system model problem. In this paper, we present a new hybrid algorithm for low-resolution image enhancement, whose parameters are automatically adjusted by the training data and, in contrast to other super-resolution methods, do not require regular adjustment parameters. The method is a hybrid method that includes self-organizing maps as a preprocessor, the k-nearest neighbor algorithm as a classifier, and the Laplace operational edge detection operator as an edge extractor. We built a single external dictionary using a combination of low-resolution and high-resolution feature patches and then train our proposed network. Subsequently, we reconstruct the high-resolution image by Converting the low-resolution input image to feature patch vectors. Then for each vector, find the matching neuron in the network and retrieve all the vectors that belong to it. Then we train the k-nearest neighbor algorithm with these vectors plus the input vector and find the best vector most similar to the input vector and reconstruct our high super-resolution image. The proposed image super-resolution method presents in practical experiments better results with better resolution and quality than many traditional and state-of-the-art methods, both visually compared with each other using human and computational benchmarks to compare the quality of the image super-resolution algorithms. The proposed image enhancement method is best for reconstructing high-resolution images that need high-frequency details and sharp edges with a smooth slope of image objects in their structures.