scholarly journals Computer Vision in Precipitation Nowcasting: Applying Image Quality Assessment Metrics for Training Deep Neural Networks

Atmosphere ◽  
2019 ◽  
Vol 10 (5) ◽  
pp. 244 ◽  
Author(s):  
Quang-Khai Tran ◽  
Sa-kwang Song
Keyword(s):  
Neural Network ◽  
Computer Vision ◽  
Image Quality ◽  
Quality Assessment ◽  
Image Quality Assessment ◽  
Mean Squared Error ◽  
Neural Network Models ◽  
Common Mean ◽  
Radar Images ◽  
Squared Error

This paper presents a viewpoint from computer vision to the radar echo extrapolation task in the precipitation nowcasting domain. Inspired by the success of some convolutional recurrent neural network models in this domain, including convolutional LSTM, convolutional GRU and trajectory GRU, we designed a new sequence-to-sequence neural network structure to leverage these models in a realistic data context. In this design, we decreased the numbers of channels in high abstract recurrent layers rather than increasing them. We formulated the task as a problem of encoding five radar images and predicting 10 steps ahead at the pixel level, and found that using only the common mean squared error can misguide the training and mislead the testing. Especially, the image quality of last predictions usually degraded rapidly. As a solution, we employed some visual image quality assessment techniques including Structural Similarity (SSIM) and multi-scale SSIM to train our models. Experimental results show that our structure was more tolerant to increasing uncertainty in the data, and the use of image quality metrics can significantly reduce the blurry image issue. Moreover, we found that using SSIM was very effective and a combination of SSIM with mean squared error and mean absolute error yielded the best prediction quality.

A Very Fast and Accurate Image Quality Assessment Method based on Mean Squared Error with Difference of Gaussians

2020 ◽  
Vol 64 (1) ◽  
pp. 10502-1-10502-5
Author(s):  
Sung-Ho Bae ◽  
Seong-Bae Park
Keyword(s):  
Image Quality ◽  
Quality Assessment ◽  
Image Quality Assessment ◽  
Mean Squared Error ◽  
Visual Quality ◽  
Prediction Performance ◽  
Difference Of Gaussians ◽  
Mathematical Properties ◽  
Squared Error ◽  
Computational Simplicity

Abstract Mean squared error (MSE) has long been the most useful objective image quality assessment (IQA) metric due to its mathematical tractability and computational simplicity, although it has shown poor correlations with the perceived visual quality for distorted images. Contrary to the MSE, recent IQA methods are more closely related with measured visual quality. However, their applications are somewhat limited due to their heavy computational costs and inapplicability in optimization process. In order to develop a better IQA method that will be closer to the perceived visual quality, the authors aimed to incorporate simple yet powerful linear features into the form of MSE while retaining the advantages of computational simplicity and desirable mathematical properties of MSE. Through comprehensive experiments, the authors found that Difference of Gaussians (DoG) kernel significantly improves the prediction performance while keeping the aforementioned advantages in the form of MSE. The proposed method performs better as the DoG filtering well approximates the behaviors of neural response functions in the visual cortex of the human visual system, thus extracting perceptually important features. At the same time, it holds the computational simplicity and mathematical properties of MSE since DoG is a very simple linear kernel. Their extensive experiments showed that the proposed method provides competitive prediction performance to the recent IQA methods with a significantly lower computational complexity.

No-Reference Image Quality Assessment Based on Multi-Order Gradients Statistics

2020 ◽  
Vol 64 (1) ◽  
pp. 10505-1-10505-16
Author(s):  
Yin Zhang ◽  
Xuehan Bai ◽  
Junhua Yan ◽  
Yongqi Xiao ◽  
C. R. Chatwin ◽  
...  
Keyword(s):  
Neural Network ◽  
Image Quality ◽  
Quality Assessment ◽  
Image Quality Assessment ◽  
Assessment Method ◽  
Image Distortion ◽  
Image Feature ◽  
Reference Image ◽  
Gradient Magnitude ◽  
Distortion Type

Abstract A new blind image quality assessment method called No-Reference Image Quality Assessment Based on Multi-Order Gradients Statistics is proposed, which is aimed at solving the problem that the existing no-reference image quality assessment methods cannot determine the type of image distortion and that the quality evaluation has poor robustness for different types of distortion. In this article, an 18-dimensional image feature vector is constructed from gradient magnitude features, relative gradient orientation features, and relative gradient magnitude features over two scales and three orders on the basis of the relationship between multi-order gradient statistics and the type and degree of image distortion. The feature matrix and distortion types of known distorted images are used to train an AdaBoost_BP neural network to determine the image distortion type; the feature matrix and subjective scores of known distorted images are used to train an AdaBoost_BP neural network to determine the image distortion degree. A series of comparative experiments were carried out using Laboratory of Image and Video Engineering (LIVE), LIVE Multiply Distorted Image Quality, Tampere Image, and Optics Remote Sensing Image databases. Experimental results show that the proposed method has high distortion type judgment accuracy and that the quality score shows good subjective consistency and robustness for all types of distortion. The performance of the proposed method is not constricted to a particular database, and the proposed method has high operational efficiency.

Precise No-Reference Image Quality Evaluation Based on Distortion Identification

2021 ◽  
Vol 17 (3s) ◽  
pp. 1-21
Author(s):  
Chenggang Yan ◽  
Tong Teng ◽  
Yutao Liu ◽  
Yongbing Zhang ◽  
Haoqian Wang ◽  
...  
Keyword(s):  
Neural Network ◽  
Image Quality ◽  
Quality Assessment ◽  
Large Scale ◽  
Quality Evaluation ◽  
Image Quality Assessment ◽  
State Of The Art ◽  
Gaussian White Noise ◽  
The State ◽  
Reference Image

The difficulty of no-reference image quality assessment (NR IQA) often lies in the lack of knowledge about the distortion in the image, which makes quality assessment blind and thus inefficient. To tackle such issue, in this article, we propose a novel scheme for precise NR IQA, which includes two successive steps, i.e., distortion identification and targeted quality evaluation. In the first step, we employ the well-known Inception-ResNet-v2 neural network to train a classifier that classifies the possible distortion in the image into the four most common distortion types, i.e., Gaussian white noise (WN), Gaussian blur (GB), jpeg compression (JPEG), and jpeg2000 compression (JP2K). Specifically, the deep neural network is trained on the large-scale Waterloo Exploration database, which ensures the robustness and high performance of distortion classification. In the second step, after determining the distortion type of the image, we then design a specific approach to quantify the image distortion level, which can estimate the image quality specially and more precisely. Extensive experiments performed on LIVE, TID2013, CSIQ, and Waterloo Exploration databases demonstrate that (1) the accuracy of our distortion classification is higher than that of the state-of-the-art distortion classification methods, and (2) the proposed NR IQA method outperforms the state-of-the-art NR IQA methods in quantifying the image quality.

Sign in / Sign up

Export Citation Format

Share Document