Abstract: Techniques for creating machine learning models for direct homography regression for image rectification are described. In certain embodiments, a training service trains an algorithm on a source view of a training image and a homography matrix of the training image into a machine learning model that generates a normalized homography matrix for an input of the source view. The normalized homography matrix may then be utilized to generate a target view of an image input into the machine learning model. The target view of the image may be used in a document processing pipeline for document images captured using cameras.

Abstract: Techniques for document rectification via homography recovery using machine learning are described. An image rectification system can intelligently make use of multiple pipelines for rectifying document images based on the detected type of device that generated the images. The image rectification system can provide high-quality rectifications without requiring human cooperation, multiple views of the document in multiple images, and/or without being constrained to only be able to process images from one source context.

Abstract: Objects detected in data may be filtered from an object recognition index. Data for object detection may be received. An object detection technique may be applied to the data to detect an object. If the object does not satisfy indexing criteria for the object recognition index, then the detected object may be excluded from the object recognition index.

Abstract: Objects detected in data may be filtered from an object recognition index. Data for object detection may be received. An object detection technique may be applied to the data to detect an object. If the object does not satisfy indexing criteria for the object recognition index, then the detected object may be excluded from the object recognition index.

Abstract: Techniques for machine learning-based image super-resolution are described. A Dual Path Deep Back Projection Network can be used to enhance an input image. For example, the model may be trained to perform image super-resolution, remove artifacts, provide filtering or low light enhancement, etc. Classification may be performed on the resulting enhanced images to identify objects represented in the images. The model may be trained using a dataset that includes groups of images: an original image and an enhanced image. The model may use both residual and dense connectivity patterns between each successive back projection blocks to improve construction of a high-resolution output image from a low resolution input image. The enhanced images increase classification accuracy for input images having low image resolution.