A Hybrid MLP-PNN Architecture for Fast Image Superresolution

Abstract

Esta es la versión de autor de la comunicación de congreso publicada en: This is an author produced version of a paper published in: Abstract. Image superresolution methods process an input image sequence of a scene to obtain a still image with increased resolution. Classical approaches to this problem involve complex iterative minimization procedures, typically with high computational costs. In this paper is proposed a novel algorithm for superresolution that enables a substantial decrease in computer load. First, a probabilistic neural network architecture is used to perform a scattered-point interpolation of the image sequence data. The network kernel function is optimally determined for this problem by a multi-layer perceptron trained on synthetic data. Network parameters dependence on sequence noise level is quantitatively analyzed. This super-sampled image is spatially filtered to correct finite pixel size effects, to yield the final high-resolution estimate. Results on a real outdoor sequence are presented, showing the quality of the proposed method.

Super-resolution: a comprehensive survey

“…Examples of such networks are Linear Associative Memories (LAM) with single [61] and dual associative learning [192], Hopfield NN [96], [326], Probabilistic NN [130], [304], Integrated Recurrent NN [136], Multi Layer Perceptron (MLP) [196], [354], [385], [547], Feed Forward NN, [232], [233], and Radius Basis Function (RBF) [327], [607].…”

“…-some of the learning based approaches perform fast in their testing phase, but they need a time consuming training step and are mostly application dependent, like in [130], [292], [304].…”

Limits of Learning-Based Superresolution Algorithms

“…Candocia and Principe (1999) use an NN to cluster LR patches as well as infer HR patches from LR patches. Miravet and Rodríguez (2003) use a hybrid multilayer perception (MLP) and probabilistic neural network (PNN) for SR via nonuniform interpolation. Kursun and Favorov (2003) build a biologically inspired NN that mimics the SINBAD (Set of INteracting BAckpropagating Dendrites) cells in the visual cortex, and claim that the SINBAD NN can identify high-order regularities in natural images.…”

“…Among them, the algorithms in Freeman and Pasztor (1999), Bégin and Ferrie (2004), Bishop et al (2003), Candocia and Principe (1999), Kursun and Favorov (2003), Miravet and Rodríguez (2003), Pickup et al (2003), Sun et al (2003), Zhang and Pan (2002) can be applied to general images or videos (Bishop et al 2003). As the image/video sizes are not fixed, all of them can only use patch-based approaches.…”

Accurate and Robust Image Superresolution by Neural Processing of Local Image Representations

“…Recently [6,7], the authors have proposed a neural network based technique that provides results comparable to classical methods with a substantial decrease in computational complexity. This technique estimates image values in a dense grid using an irregular interpolation scheme, with distance dependent interpolation weights.…”