Robust image retrieval by cascading a deep quality assessment network

作者：

Highlights：

• Image enhancement for robust image retrieval (IR) in the presence of distortions

• Quality assessment for image retrieval (QAIR) convolutional neural network (CNN) predicts image retrieval performance of a distorted image in terms of average precision

• An enhancement CNN is optimized for image retrieval using the QAIR CNN

• Experiments with noise and low resolution on multiple datasets reveal performance improvements

• Method to adapt deep learning features for IR that are robust to distortions

摘要

•Image enhancement for robust image retrieval (IR) in the presence of distortions•Quality assessment for image retrieval (QAIR) convolutional neural network (CNN) predicts image retrieval performance of a distorted image in terms of average precision•An enhancement CNN is optimized for image retrieval using the QAIR CNN•Experiments with noise and low resolution on multiple datasets reveal performance improvements•Method to adapt deep learning features for IR that are robust to distortions

论文关键词：Image enhancement,Image quality assessment,Deep convolutional neural network,Denoising,Super resolution,Image retrieval

论文评审过程：Received 7 February 2019, Revised 28 September 2019, Accepted 28 September 2019, Available online 5 October 2019, Version of Record 8 October 2019.

论文官网地址：https://doi.org/10.1016/j.image.2019.115652