Abstract: A device such as a color printer includes a main memory, a cache memory, and a convolutional neural network configured to convert pixels from a first color space to a second color space. The convolutional neural network is organized into execution-separable layers, and loaded one or more layer at a time (depending on cache size) from the main memory to the cache memory, whereby the pixels are processed through each of the layers in the cache memory, and layers that have completed processing are evicted to make room for caching next layer(s) of the network.

Abstract: A method includes generating multiple attention map for an image from outputs of a corresponding different one of multiple convolutional hidden layers of a neural network. A different weighted attention map is then generated from each of the attention map. The weighted attention maps are input to a first fully-connected neural network layer to generate a colorfulness metric, which may be used to augment human-perceived colorfulness of the image.

Abstract: Embodiments provide an automated approach for generating unbiased synthesized image-label pairs for colorization training of retro photographs. Modern grayscale images with corresponding color images are translated to images with the characteristics of retro photographs, thereby producing training data that pairs images with the characteristics of retro paragraphs with corresponding color images. This training data can then be employed to train a deep learning model to colorize retro photographs more effectively.

Abstract: A digital image comparator includes an Faster R-CNN configured to generate a filtered set of local feature maps of an input image, a match head, and logic to preserve the local feature maps in an indexed data structure and to make the local feature maps retrievable by the match head via local feature map indexes.

Abstract: A scanning control for a photocopier includes a pair of identically weighted networks configured to perform feature extraction on images and a memory storing registered security patterns. A match head of the scanning control receives an image pair, wherein a first image of the image pair is generated by a scanning element of the photocopier, and a second image of the image pair is obtained from the registered security patterns, and outputs a match score for the image pair. The output of the match head coupled controls operation of the scanning element.

Abstract: A method of detecting a digital stamp pattern involves operating a scanning device to create a page image scanned from a document page, inputting the page image into a one-shot trained neural network, the one-shot trained neural network configured to recognize a copy-guard digital stamp pattern using one-shot learning, analyzing the page image using the one-shot trained neural network to detect the copy-guard digital stamp pattern, and on condition the copy-guard digital stamp pattern is detected, issuing an electronic alert.

Abstract: A digital image processor includes a region proposal network configured to transform digital image inputs into region proposals and bounding box refinement logic configured to transform the region proposals by determining a first set of the region proposals exhibiting dense spacing, determining a second set of the region proposals exhibiting sparse spacing, executing a first transformation to merge at least some of the region proposals exhibiting dense spacing to generate refined region proposals, executing a second transformation to join at least some of the region proposals exhibiting sparse spacing to generate additional ones of the refined region proposals, and applying an expansion transformation to the refined region proposals.

Abstract: Systems and methods upscale an input image by a final upscaling factor. The systems and methods employ a first module implementing a super resolution neural network with feature extraction layers and multiple sets of upscaling layers sharing the feature extraction layers. The multiple sets of upscaling layers upscale the input image according to different respective upscaling factors to produce respective first module outputs. The systems and methods select the first module output with the respective upscaling factor closest to the final upscaling factor. If the respective upscaling factor for the selected first module output is equal to the final upscaling factor, the systems and methods output the selected first module output. Otherwise, the systems and methods provide the selected first module output to a second module that upscales the selected first module output to produce a second module output corresponding to the input image upscaled by the final upscaling factor.

Abstract: A method includes generating multiple attention map for an image from outputs of a corresponding different one of multiple convolutional hidden layers of a neural network. A different weighted attention map is then generated from each of the attention map. The weighted attention maps are input to a first fully-connected neural network layer to generate a colorfulness metric, which may be used to augment human-perceived colorfulness of the image.

Abstract: A method and apparatus for detecting a digital stamp pattern are disclosed. Keypoints and descriptors are extracted from an original template pattern image. A low resolution original document and at least one lower resolution template pattern image are template-matched to detect a matched region based on match correlation coefficients. This region is cropped out of a full resolution original document. Keypoints and descriptors are extracted from the cropped region, and are matched with stamp pattern keypoints and descriptors using feature based pattern matching. A transformation matrix is used to detect scaling, rotation, and translation of a detected digital stamp pattern in the cropped region. A number of qualified matches determined using feature based pattern matching or the transformation matrix are checked against a pre-set threshold. If a pre-set threshold is exceeded, an alert is generated for a possible security issue. Otherwise, a no security issues signal may be generated.

Abstract: Embodiments provide an automated approach for generating unbiased synthesized image-label pairs for colorization training of retro photographs. Modern grayscale images with corresponding color images are translated to images with the characteristics of retro photographs, thereby producing training data that pairs images with the characteristics of retro paragraphs with corresponding color images. This training data can then be employed to train a deep learning model to colorize retro photographs more effectively.

Abstract: Systems and methods for processing images receive an input image. The systems and methods provide the input image to a first module to increase a resolution of the input image to produce an upscaled image. The systems and methods detect white pixels in the input image. The systems and methods generate a mask associated with the input image. The mask includes mask bits that are set to mark the white pixels in the input image. The systems and methods upscale the mask to produce an upscaled mask matching a resolution of the upscaled image. The systems and methods identify target pixels of the upscaled image that correspond to the set mask bits in the upscaled mask. The systems and methods modify the upscaled image to produce an output image by replacing target pixels of the upscaled image with a replacement pixel having greater whiteness. The systems and methods output the output image.

Abstract: Methods and systems for training and utilizing an artificial neural network (ANN) are provided. In an example method, a computing device could receive an input image comprising a plurality of channels and determine a saliency map for the input image. The computing device could also establish at least one of the plurality of channels as a training channel and at least some of the plurality of channels as one or more ground truth channels. Further, the computing device could train an ANN to predict one or more output channels from the one or more training channels, where the training involves computationally updating weights of the ANN based on a loss function that comprises a difference between the one or more output channels and the one or more ground truth channels, and where the difference is computationally biased based on values from the saliency map.

Abstract: Systems and methods for processing images receive an input image. The systems and methods provide the input image to a first module to increase a resolution of the input image to produce an upscaled image. The systems and methods detect white pixels in the input image. The systems and methods generate a mask associated with the input image. The mask includes mask bits that are set to mark the white pixels in the input image. The systems and methods upscale the mask to produce an upscaled mask matching a resolution of the upscaled image. The systems and methods identify target pixels of the upscaled image that correspond to the set mask bits in the upscaled mask. The systems and methods modify the upscaled image to produce an output image by replacing target pixels of the upscaled image with a replacement pixel having greater whiteness. The systems and methods output the output image.

Abstract: Systems and methods upscale an input image by a final upscaling factor. The systems and methods employ a first module implementing a super resolution neural network with feature extraction layers and multiple sets of upscaling layers sharing the feature extraction layers. The multiple sets of upscaling layers upscale the input image according to different respective upscaling factors to produce respective first module outputs. The systems and methods select the first module output with the respective upscaling factor closest to the final upscaling factor. If the respective upscaling factor for the selected first module output is equal to the final upscaling factor, the systems and methods output the selected first module output. Otherwise, the systems and methods provide the selected first module output to a second module that upscales the selected first module output to produce a second module output corresponding to the input image upscaled by the final upscaling factor.

Abstract: An example system includes a processor and a non-transitory computer-readable medium having stored therein instructions that are executable to cause the system to perform various functions. The functions include obtaining an image associated with a print job, and providing the image as input to a convolutional neural network. The convolutional neural network includes a residual network, upscaling layers, and classification layers configured to detect whether the image is an artificial image having a computer-generated image gradient. The functions also include determining, based on an output of the classification layers, that the image is an artificial image having a computer-generated image gradient. Further, the functions include, based on determining that the image is an artificial image having a computer-generated image gradient, providing the image to an upscaling module of a print pipeline for upscaling rather than using an output of the upscaling layers for the upscaling.

Abstract: Methods and systems for training and utilizing an artificial neural network (ANN) are provided. In an example method, a computing device can receive an image pair, where a first image of the image pair includes a training image and a second image of the image pair includes a ground truth image. The computing device can provide instances of the first image to a plurality of image filtering modules and determine respective filtered representations of the first image using the plurality of image filtering modules. The computing device can indirectly train an adaptor ANN by applying the adaptor ANN on the respective filtered representations to produce an adapted representation; determining, using a trained colorization ANN, a colorized image from the adapted representation; and updating weights of the adaptor ANN based on a loss function that comprises a difference between the colorized image and the second image of the image pair.

Abstract: Methods and systems for training and utilizing an artificial neural network (ANN) are provided. In an example method, a computing device can receive an image pair, where a first image of the image pair includes a training image and a second image of the image pair includes a ground truth image. The computing device can provide instances of the first image to a plurality of image filtering modules and determine respective filtered representations of the first image using the plurality of image filtering modules. The computing device can indirectly train an adaptor ANN by applying the adaptor ANN on the respective filtered representations to produce an adapted representation; determining, using a trained colorization ANN, a colorized image from the adapted representation; and updating weights of the adaptor ANN based on a loss function that comprises a difference between the colorized image and the second image of the image pair.

Abstract: Methods and systems for training and utilizing an artificial neural network (ANN) are provided. In an example method, a computing device could receive an input image comprising a plurality of channels and determine a saliency map for the input image. The computing device could also establish at least one of the plurality of channels as a training channel and at least some of the plurality of channels as one or more ground truth channels. Further, the computing device could train an ANN to predict one or more output channels from the one or more training channels, where the training involves computationally updating weights of the ANN based on a loss function that comprises a difference between the one or more output channels and the one or more ground truth channels, and where the difference is computationally biased based on values from the saliency map.

Abstract: Methods and systems for training and utilizing an artificial neural network (ANN) are provided. In an example method, a computing device can receive an image pair, where a first image of the image pair includes a training image and a second image of the image pair includes a ground truth image. The computing device could utilize a trained de-noise ANN to determine a de-noised representation of the first image. The computing device could then indirectly training an adaptor ANN by at least applying the adaptor ANN on the de-noised representation to produce an adapted representation for the first image; determining, using a trained super resolution ANN, a high resolution image from the adapted representation, and computationally updating weights of the adaptor ANN based on a loss function that comprises a difference between the high resolution image and the second image.