Image for post
Image for post

Image Quality Assessment: A Survey

Introduction

Image Distortions

Image for post
Image for post
Image for post
Image for post
Image for post
Image for post
Fig. 1. (a) reference image, (b) JP2K compression, (c) gaussian blur
Image for post
Image for post
Image for post
Image for post
Image for post
Image for post
Fig. 2. (a) reference image, (b) JPEG compression, (c) white noise.

Literature Review

Problem Statement

Methodologies

Deep CNN-Based Blind Image Quality Predictor (DIQA)

Image for post
Image for post
Fig. 3. Overall flowchart of DIQA. Source: http://bit.ly/2Ldw4PZ
Image for post
Image for post
Fig. 4. The architecture for the objective error map prediction. The red and blue arrows indicate the flows of the first and stage. Source: http://bit.ly/2Ldw4PZ
Image for post
Image for post
The loss function for stage 1, where functions g and f are defined in Fig. 4.
Image for post
Image for post
Image for post
Image for post
Image for post
Image for post
The loss function for stage 2, where mu and sigma are handcrafted features, and S is the subjective score (MOS).

Blind Image Quality Estimation via Distortion Aggravation (BMPRI)

Image for post
Image for post
Blocking effect where Q controls the compression quality.
Image for post
Image for post
Ringing effect where R controls the compression ratio.
Image for post
Image for post
Blurring effect where g is a Gaussian kernel and * is a convolution operator.
Image for post
Image for post
Noising effect where N(0, v) generates normally distributed random values with 0 mean and v variance.
Image for post
Image for post
The formula for LBP calculation where g_p and g_c are pixels and P, R denote the neighbor number and radius of the LBP structure.
Image for post
Image for post
Image for post
Image for post
The overlap between the distorted image and MPRI feature maps.
Image for post
Image for post
LBP feature map where c is set to different values for the given distortion types. For example, c is set to 0 or 1 to estimate noising effects.
Image for post
Image for post
The score assigned to a given aggravation m where high q means a worse quality.
Image for post
Image for post

Blind Image Quality Assessment Based on High Order Statistics Aggregation (HOSA)

Image for post
Image for post
Fig. 5. The pipeline of the proposed method HOSA. Source: lampsrv02.umiacs.umd.edu/projdb/project.php?id=76
Image for post
Image for post
The resulting vector of normalized patches for each image. Each patch is a matrix of B x B dimensions. A whitening process is performed to remove linear correlations between patches.
Image for post
Image for post
The cumulative approximation error.
Image for post
Image for post
The mean, covariance,ƒ and coskewness of each cluster k.
Image for post
Image for post
Residuals between the soft weighted mean of local features for cluster k and the mean of cluster k where d denotes the d-th dimension of a vector
Image for post
Image for post
The Gaussian kernel similarity weight between local feature x and codeword k.
Image for post
Image for post
Image for post
Image for post
Soft weighted second and third-order statistics between local features and codewords.
Image for post
Image for post

Performance Comparison

Image for post
Image for post
Fig. 7. The SRCC results published by the method’s authors based on the CSIQ dataset.

Jupyter Notebook

Conclusion

References

Data Scientist, I research and blog about machine learning in my spare time

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store