The invention discloses a learning method for joint low-rank representation and sparse regression. The method comprises the following steps: extracting the features of a SUN data set with an image memory score label; putting low-rank representation and a sparse regression model under the same framework to form a whole, and constructing a joint low-rank representation and sparse regression model; and solving the automatic image prediction memorability problem using a multi-vision adaptive regression algorithm, getting the relationship between image features and image memory under optimal parameters, getting a relationship result under optimal parameters, predicting the image memory of a test set in a database, and verifying the prediction result based on relevant evaluation criteria. The memorability of an image area can be accurately predicted through a low-rank learning framework of joint low-rank representation and sparse regression.