PATCH SELECTION FOR NEURAL NETWORK BASED NO-REFERENCE IMAGE QUALITY ASSESSMENT

• Main Authors: XU, Xuyuan, ZHOU, Chang, WONG, Hon Wah, PO, Lai Man, LUK, Hon Tung, LI, Yuming, YUEN, Yiu Fai, CHEUNG, Hok Kwan, LAU, Kin Wai, LIU, Mengyang
• Format: Patent
• Language: English
• Published: 28.11.2019
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; IMAGE DATA PROCESSING OR GENERATION, IN GENERAL; PHYSICS

The present disclosure relates to a method for image patch selection for training a neural network for image quality assessment. The method includes receiving an input image and extracting one or more image patches from the input image. The moment of the extracted image patches is measured. There is a decision to accept or decline the extracted image patches according to the measured moment. Additional image patches are extracted until a minimum number, Nmin, of extracted image patches are accepted. Alternatively, selection criteria are adjusted until the minimum number of extracted image patches are accepted. The selected image patches are input into a neural network with a corresponding image quality value of the input image, and the neural network is trained with the image patches and image quality value. Also provided is a method for image quality assessment using a neural network trained as set forth above.