TY - JOUR
T1 - No-reference fundus image quality assessment using convolutional neural network
AU - Bhatkalkar, Bhargav J.
AU - Reddy, Dheeraj Rajaram
AU - Dasu, Vishnu Asutosh
AU - Raykar, Advait
AU - Prabhu, Srikanth
AU - Bhandary, Sulatha
AU - Hegde, Govardhan
PY - 2019/4/1
Y1 - 2019/4/1
N2 - Computerized fundus image analysis is a well-established research area in the field of medical imaging. The cause of noise in fundus images is due to many factors like the low lighting conditions, adverse illumination effects, camera malfunctioning, etc. The presence of noise may lead to data loss and sometimes to the wrong data interpretation. Classifying the fundus images into either good quality or bad quality is very important as the good quality fundus images can be directly sent for processing without any preprocessing, hence reducing the computational time and the bad quality images can be forwarded for the required preprocessing stages. In this paper, we are using a convolutional neural network (CNN) to assess the quality of fundus images automatically. We use No-reference image quality assessment technique (IQA) classify the fundus images into good quality or bad quality based on their quality. A Mean Opinion Square (MOS) of 12 image quality assessment participants is taken for labeling the 300 fundus images based on their quality. The participants have rated the fundus images on the scale of 0-10, where the 0-rating is given for very bad quality fundus images, and 10-rating is given for the very good quality fundus images. The experimental study has proven that the classification result of the proposed CNN outperforms the best-known blind image quality assessment algorithms, namely, DIVINE, BLIINDS-II, and BRISQUE when trained on the public databases LIVE, TID2013 and on our fundus image dataset.
AB - Computerized fundus image analysis is a well-established research area in the field of medical imaging. The cause of noise in fundus images is due to many factors like the low lighting conditions, adverse illumination effects, camera malfunctioning, etc. The presence of noise may lead to data loss and sometimes to the wrong data interpretation. Classifying the fundus images into either good quality or bad quality is very important as the good quality fundus images can be directly sent for processing without any preprocessing, hence reducing the computational time and the bad quality images can be forwarded for the required preprocessing stages. In this paper, we are using a convolutional neural network (CNN) to assess the quality of fundus images automatically. We use No-reference image quality assessment technique (IQA) classify the fundus images into good quality or bad quality based on their quality. A Mean Opinion Square (MOS) of 12 image quality assessment participants is taken for labeling the 300 fundus images based on their quality. The participants have rated the fundus images on the scale of 0-10, where the 0-rating is given for very bad quality fundus images, and 10-rating is given for the very good quality fundus images. The experimental study has proven that the classification result of the proposed CNN outperforms the best-known blind image quality assessment algorithms, namely, DIVINE, BLIINDS-II, and BRISQUE when trained on the public databases LIVE, TID2013 and on our fundus image dataset.
UR - http://www.scopus.com/inward/record.url?scp=85069742031&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85069742031&partnerID=8YFLogxK
M3 - Article
AN - SCOPUS:85069742031
SN - 2277-3878
VL - 7
SP - 663
EP - 667
JO - International Journal of Recent Technology and Engineering
JF - International Journal of Recent Technology and Engineering
IS - 6
ER -