Page 142 - Computational Retinal Image Analysis

136 CHAPTER 8 Image quality assessment

1.1 Image quality of ophthalmic images
The quality of an image is an important factor that affects image and video analysis
systems across a range of different applications including image capture, compression,
transmission, segmentation and registration [1]. The reliability of automated algorithms
applied to images are linked very closely to the range of image quality that they operate
upon. Ophthalmic images can vary in quality for a variety of reasons. These are
sometimes related to the ocular health of the patient. For example, in the case of retinal
imaging, poor image quality can be caused by media opacity such as cataract, vitreous
hemorrhages, asteroid hyalosis, etc. Variation in pupil size (either as a consequence of
dilation or natural variation) can affect image quality [2]. Changes in the quality of an
image can also be caused by differences in operator behaviour at the time of image capture
where differences in camera exposure and focal plane error can all lead to differences in
quality. Artifacts relating to the lens of the image capture system (such as debris on the
lens) can also lead to changes in image quality. The subject may blink, therefore leading
to eye-lashes being inadvertently included in the image. Also, the subject may move
their head or eyes, leading to blurring of the image. Image compression techniques can
also affect the quality of an image. The factors described above affect the appearance of
an image in terms of low contrast, poor focus (e.g. blurring), dark areas on the image and
artifacts on the image. Some examples of different retinal fundus images showing how
image quality is affected in different ways are shown in Fig. 1.
Algorithms that correct features of poor image quality, such as correcting for
uneven illumination and increasing contrast [4, 5], can improve image quality in
cases where specific image capture problems arise. Algorithms that evaluate image
quality have the potential to alert operators to image clarity problems very soon after
the original image has been captured and the subject is still in-situ. A further image
can be captured to improve the clarity of the image with minimum inconvenience
to the operator or subject. The judgment of the quality of an image relates to how
the image will be used in the context of different clinical purposes. IQA algorithms
reflect the requirement of a range of different applications; examples of these are
discussed in the following section.

1.2 Applications of image quality assessment algorithms
The judgment of whether an image is high quality or low quality is highly dependent
on the application that the image is intended for. For example, some images are
required for inclusion in ophthalmic screening systems while others may contribute
to an epidemiological study requiring morphometric measurements of retinal vessels.
Both applications require different aspects of image quality to be evaluated. For
screening systems, clarity of the entire image is required to ensure signs of pathology
are not missed. In order to obtain reliable vessel morphometric measurements
however, image clarity should be adequate to allow accurate vessel segmentation for
at least a portion of the image [6]. Human observers judge the quality of images in
different ways depending on the application, and automated IQA algorithms reflect