This is a demonstration of the algorithm described in the paper "GENERATING ADAPTIVE AND ROBUST FILTER SETS USING AN UNSUPERVISED LEARNING FRAMEWORK". The code provides a method to control the correlation property of weights when learning a sparse autoencoder.
You can change this program as you like and use it anywhere, but please refer to its original source (cite our paper and our web page at https://ghassanalregib.com/unique-unsupervised-image-quality-estimation-2/).
Code for both Image Quality Assessment and Texture Retrieval are provided in separate folders.
For each application, run the demo.m file. Sample original and distorted images (for IQA) and sample texture images are provided in separate folders (iqa and texture). Please copy these images into the main IQA and Texture folders and set the path accordingly in each demo.m file.
For IQA : The code outputs the quality of the distorted image which lies between 0 and 1 (1 being the best quality).
For Texture Retrieval : The code calculates the similarity between the provided 2 texture images between 0 and 1 (1 being the most similar).
In this paper, we introduce an adaptive unsupervised learning framework, which utilizes natural images to train filter sets. The applicability of these filter sets is demonstrated by evaluating their performance in two contrasting applications - image quality assessment and texture retrieval. While assessing image quality, the filters need to capture perceptual differences based on dissimilarities between a reference image and its distorted version. In texture retrieval, the filters need to assess similarity between texture images to retrieve closest matching textures. Based on experiments, we show that the filter responses span a set in which a monotonicity-based metric can measure both the perceptual dissimilarity of natural images and the similarity of texture images. In addition, we corrupt the images in the test set and demonstrate that the proposed method leads to robust and reliable retrieval performance compared to existing methods.
Demonstrating the controlled whitening process of data by visualizing covariance matrices and their corresponding learnt weights :
The results of the proposed method for IQA on LIVE, and TID13 databases :
The results of the proposed method for Texture Retrieval on CUReT database :
The robustness of the method evaluated on texture retrieval when AWGN is added to the texture images :
IEEE Link : https://ieeexplore.ieee.org/document/8296841
ArXiv Link : https://arxiv.org/abs/1811.08927