Blind image deblurring with cascade architecture

One or more computer storage media with device-executable instructions comprising:

• device-executable instructions to receive, at a processor, a blurred image obtained by an image capture device;
• device-executable instructions to apply the blurred image to a first blur predictor or estimator of a plurality of blur predictors or estimators to generate a first blur kernel or recovered image;
• device-executable instructions to apply the first blur kernel or recovered image to a first trained machine learning predictor of a plurality of trained machine learning predictors to obtain a first restored version of the blurred image;
• device-executable instructions to apply the first restored version of the blurred image to a subsequent blur predictor or estimator of the plurality of blur predictors or estimators to generate a subsequent blur kernel or recovered image, the plurality of blur predictors or estimators and the plurality of trained machine learning predictors being arranged in a cascade architecture comprising a plurality of layers connected in series, individual layers of the plurality of layers comprising at least one of the trained machine learning predictors and at least one of the plurality of blur predictors or estimators, and
• device-executable instructions to use the cascade architecture to interleave steps of calculating revised estimates from the restored versions with trained machine learning prediction steps.

An image deblur engine comprising:

• a processor arranged to receive a blurred image obtained by an image capture device;
• the processor further arranged to apply the blurred image to a first blur predictor or estimator of a plurality of blur predictors or estimators to generate a first blur kernel or recovered image;
• the processor further arranged to apply the first blur kernel or recovered image to a first trained machine learning predictor of a plurality of trained machine learning predictors to calculate a first restored version of the blurred image;
• the processor further arranged to apply the first restored version of the blurred image to a subsequent blur predictor or estimator of the plurality of blur predictors or estimators to generate a subsequent blur kernel or recovered image, the plurality of blur predictors or estimators and the plurality of trained machine learning predictors being arranged in a cascade architecture comprising a plurality of layers connected in series, individual layers of the plurality of layers comprising at least one of the trained machine learning predictors and at least one of the plurality of blur predictors or estimators, and to use the cascade architecture to interleave steps of calculating revised estimates from the restored versions with trained machine learning prediction steps.