Research On Image Memorability Prediction Method Based On Low-rank Representation Learning

Nowadyas,social network has become an indispensable part of people's life.As the main contents of the generated contents(UGCs),images are everywhere.We have a glance of millions of images everyday.Some of them are remembered for a long time while some are forgotten after only a glance.It has been proved that memorability is an intrinsically stable property of images which measures the degree to which or how an image is remembered.It can be applied in fields of image processing,computer vision,user interface design,video summarization and advertisement design.Many studies have explored the potential relevance between image content and image memorability.However,the research on automatic prediction of image memorability by image features has not been carried out in large numbers.In recent years,because of good revelation of the global structural information on the spatial distribution of data and good robustness to noise,low rank learning has been widely used in the fields of multimedia and computer vision.This paper focuses on the study of image memory prediction based on low rank token learning from two aspects.: 1)Towards a sparse low-rank regression model for memorability prediction of images: the improved low rank representation and sparse regression are used to obtain the correlation between the low rank representation of image features and the image memorability,which can realize automatic feature selection,reduce feature redundancy,remove noise and improve prediction accuracy;2)A multi-view sparse low-rank regression model for memorability prediction of images: with related knowledge of multi-view feature,the accuracy of image memorability prediction can be improved by using the different perspectives and correlations of the low-level visual feature and the high-level attributes feature.We also develop an alternating direction algorithm by applying augmented Lagrangian multipliers method to solve the objective function of our methods.Experiments conducted on publicly available memorability datasets demonstrate the effectiveness of the proposed methods.