Image and Video Compression

137

7

Image Quality Assessment

With the rapid growth in Internet technologies and the enormous use of

multimedia content for different applications, from education to entertain-

ment, paramount importance is given to the use of images and videos. In

order to save space and transfer in less time, these images and videos are

compressed. Due to compression, artifacts are generated, which degrades

the quality of the images. Research is still going on to increase the compres-

sion ratio by degrading quality to a small extent. The extent to which quality

is degraded should be analyzed for further performance improvement of

the compression algorithm. The full-reference quality metrics such as peak

signal-to-noise ratio, mean square error, structural similarity, and image

delity are explained. If the reference image is not available, the technique

used to determine quality is discussed. The image quality assessment tech-

nique based on information from the reference image (reduced reference) is

also discussed.

7.1 Introduction

During the last three decades, because of advances in imaging technolo-

gies and the rapid growth of the Internet, digital images and videos are now

widely used for representing information in entertainment. We have studied

in previous chapters that due to the large size of images and videos, they

need to be compressed. As a result of the compression artifacts, the quality

of images degrades. The quality also suffers during acquisition, transmis-

sion, and reproduction and processing. So, image quality measurement has a

wide importance in many applications related to image processing. Efcient

and automatic objective quality evaluation using the development of quality

assessment algorithms is the main goal of quality assessment research. Also,

it should be consistent with subjective/human assessment. Identication

and quantication of image quality degradations by the algorithms for the

assessment are also necessary. This will be to update the performance of the

compression process or channel.

138 Image and Video Compression

Human beings are the nal recipients in many image-processing applica-

tions; images are assessed by subjects (observers). This method is based on

the mean of the opinion given by observers. Even though correct judgment

is possible, this method is fairly expensive and slow. This approach is called

subjective image quality assessment (IQA).

Predication of the perceived image quality accurately based on computa-

tional models is another way. The prediction is supposed to be in correlation

with subjective assessment. This approach is called an objective IQA.

Thus, image quality assessment techniques are classied mainly into (1)

subjective and (2) objective. Further objective quality assessment is classied

into the following:

1. Full reference (FR)—In this type of assessment, the original image

before processing (compression) is used as reference. The decom-

pressed or distorted image is compared with reference, typically

using distance criteria. This method is still used by the researchers

to evaluate performance of the compression scheme developed.

2. No reference (NR)—In this approach, there is an image available as

reference. Certain artifacts/features in an image like blockiness and

ringing are measured and given to a well-trained prediction model

to get the quality score. The model uses subjective assessment dur-

ing training. The task is difcult but is drawing the attention of

many researchers to develop no reference quality metrics.

3. Reduced reference (RR)—Certain features like edge information and

pixel values at a xed position are extracted and attached as side

information along with the compressed image. For assessment, the

features are compared and reduced so that a reference quality met-

ric can be developed. This approach can also be used for image qual-

ity improvement (repair).

Subjective IQA is discussed in the next sections, and all objective IQA

techniques are discussed in further sections along with MATLAB programs

and comparison between various quality metrics.

7.2 Subjective Image Quality Analysis

Subjective image quality analysis may be the only best method for quantify-

ing visual image quality. The mean opinion score (MOS) is a widely used

method for subjective assessment of image quality. In this method, ve

or more grade scales are used. Experts and nonexperts are asked to give

grades for images as 5 for “Excellent,” 4 for “very good,” 3 for “good,” 2 for

“bad,” and 1 for “very bad.” The average of the grades given to that image is