Visual Perceptual Quality Assessment Based on Blind Machine Learning Techniques

This paper presents the construction of a new objective method for estimation of visual perceiving quality. The proposal provides an assessment of image quality without the need for a reference image or a specific distortion assumption. Two main processes have been used to build our models: The first one uses deep learning with a convolutional neural network process, without any preprocessing. The second objective visual quality is computed by pooling several image features extracted from different concepts: the natural scene statistic in the spatial domain, the gradient magnitude, the Laplacian of Gaussian, as well as the spectral and spatial entropies. The features extracted from the image file are used as the input of machine learning techniques to build the models that are used to estimate the visual quality level of any image. For the machine learning training phase, two main processes are proposed: The first proposed process consists of a direct learning using all the selected features in only one training phase, named direct learning blind visual quality assessment DLBQA. The second process is an indirect learning and consists of two training phases, named indirect learning blind visual quality assessment ILBQA. This second process includes an additional phase of construction of intermediary metrics used for the construction of the prediction model. The produced models are evaluated on many benchmarks image databases as TID2013, LIVE, and LIVE in the wild image quality challenge. The experimental results demonstrate that the proposed models produce the best visual perception quality prediction, compared to the state-of-the-art models. The proposed models have been implemented on an FPGA platform to demonstrate the feasibility of integrating the proposed solution on an image sensor.

КОМБІНОВАНА МЕТРИКА ВІЗУАЛЬНОЇ ЯКОСТІ ЗОБРАЖЕНЬ ДИСТАНЦІЙНОГО ЗОНДУВАННЯ НА ОСНОВІ НЕЙРОННОЇ МЕРЕЖІ

The wide distribution of images of remote sensing (RS) of the Earth in various application areas makes it important to ensure the high quality of such images, which is important to identify necessary information. The complexity of the systems and the impact of various physical processes cause a significant number of distortions that lead to image corruption and possible loss of information. The use of processing methods that should reduce the impact of such factors requires control of their work, which uses quantitative indicators of visual quality. The article considers the task of creating a combined visual quality metric based on an artificial neural network (ANN), which provides high accuracy of visual quality assessment and stability of work on the noise characteristic of the RS. The problem of analysis of RS distortions is considered and the approach of using the database of test images TID2013 for verification on typical RS distortions is offered. The analysis of well-known visual quality metrics and their suitability for the estimation of such images is carried out. According to its results, it was determined that the best metrics provide the accuracy of image quality assessment for RS tasks at the level of 0.93 according to Spearman's rank-order correlation coefficient with subjective estimates of the TID2013 image database. The joint application of existing quality metrics allows eliminating the shortcomings of each of them and increasing the overall efficiency, so the article considers the problems and defines the requirements for creating a combined metric involving a neural network. A method of limiting the number of involved quality metrics with the involvement of Lasso regularization is proposed, which allows determining the most informative features (quality metrics) and simplifying the procedure of selection and reduction of their number. A study was conducted on the influence of the metric selection criterion and quantity on the accuracy of the combined metric. The influence of the structure of the neural network, the number of hidden layers, and the number of neurons in them are also analyzed. Based on the obtained results, the best implementation of ANN was selected, which with the involvement of 16 visual quality metrics allows achieving the accuracy of visual quality assessment at 0.97 according to Spearman's correlation with subjective estimates of the TID2013 database.