The present invention relates generally to digital watermarking. One aspect of the disclosure includes a method comprising: obtaining data representing imagery; using one or more configured processors, analyzing a plurality of portions of the data to detect a redundantly embedded watermark signal, the analyzing producing detection statistics for each of the plurality of portions, the detection statistics comprising a payload signature, a rotation angle and a scale factor for each portion of the plurality of portions; accumulating payload signatures based on compatible rotation angles and scale factors, said accumulating yielding an accumulated payload signature; and decoding a plural-bit payload from the accumulated payload signature. Of course, many other aspects and disclosure are provided in this patent document.

Techniques are disclosed to identify a form document in an image using a digital fingerprint of the form document. To do so, the image is evaluated to detect features of the image and generate a boundary around each feature. For each boundary, dimensions of the boundary may be stored in a color channel of a pixel in a second image. Thus, the color of the pixel represents the size of the boundary. The second image is the digital fingerprint of the form. To identify the form corresponding to the digital fingerprint, the digital fingerprint may be compared to digital fingerprints of known forms.

A decoding device may include a processor that may include a core component configured to: analyze watermark metadata to identify a watermark ROI from among multiple candidate ROIs in response to generation of the watermark metadata, determine whether rectification is to be performed within the watermark ROI, perform watermark decoding with the rectified watermark ROI data to decode data encoded within a digital watermark within the rectified watermark ROI in response to performance of the rectification, and transmit the data to a server via a network in response to successful decode of the data. The processor may also include a SIMD component configured to perform at least one of: a watermark transform with the captured image to generate the watermark metadata, and the rectification within the watermark ROI to generate the rectified watermark ROI data in response to the determination by the core component to perform the rectification.

A document verification system may verify modification to a physical document through digital image processing. Imaging sensors capture images of a modified physical document. A digital image processing circuit measures pixel parameters for a plurality of regions of the digital images for comparison to a verification pattern to verify the document.

A method for identifying whether a standard picture contains a watermark is provided. After obtaining a set of sample standard pictures, one or more sample pictures in the set of sample standard pictures are adjusted to a preset size. The sample pictures in the set of sample standard pictures do not contain watermark information. Next, an average of pixel attribute values of the sample pictures at pixel positions of the preset size is calculated. The average of the pixel attribute values at the pixel positions of the preset size is normalized to obtain the watermark-presence probabilities of the pixel positions of the preset size. Then, a target picture is adjusted to the preset size and a sum of products of pixel attribute values of the target picture at the pixel positions of the preset size and the corresponding watermark-presence probability are calculated. Finally, it is determined whether the target picture contains a watermark according to the sum of products.

A method of embedding a pattern as a watermark into a content segment. Prior to modifying the content segment, an impulse response of a filter to be used for detecting the pattern is determined; the time-reversed impulse response of the filter is inserted into the segment as the set of imperceptible features; wherein the filter is an infinite impulse response filter having a semi-white frequency spectrum and provides a pseudo-random time-domain response.

A data-bearing image (391) is created from a carrier image (371). The carrier image (371) is scaled to produce a scaled image. A clustered-dot halftone screen is applied to the scaled image to produce a halftone image. A resulting number of cells in the halftone image conforms to a cell count (372) that includes a horizontal cell value and a vertical cell value. Payload data is encoded into the halftone image to produce a data-bearing halftone image, including shifting pixel clusters within cells of the halftone image that include pixel clusters.

First shapes are removed from a first area of intersecting line patterns, and a first security pattern is added to only the first shapes. Second shapes are removed from the first area and the first security pattern, and a second security pattern is added to only the second shapes. The first and second security patterns include the intersecting line patterns phase shifted along first and second vectors. Overlap areas where the first shapes and the second shapes overlap are removed, and a third security pattern is added to only the overlap areas. The third security pattern includes altered first parallel lines and altered second parallel lines of the intersecting line patterns. A screen in a first orientation reveals a distinction between the first area and the first shapes, and the screen rotated reveals a distinction between the first area and the second shapes.

A digital watermarking system and method are disclosed. In one respect, the disclosed digital watermarking includes generating an extracted signal by applying a watermark extractor to an original image, generating a mixed signal by mixing the first signal with a periodic watermark signal using a local weighting factor for the periodic watermark signal that attenuates a strength of the watermark signal in proportion to a pixel luminance level, and replacing the extracted signal in the original image with the mixed signal to generate a marked image, wherein the watermark signal is extractable from the marked image using the watermark extractor.

Embodiments of the present invention are directed to providing new systems and methods for using deep learning techniques to generate embeddings for high dimensional data objects that can both simulate prior art embedding algorithms and also provide superior performance compared to the prior art methods. Deep learning techniques used by embodiments of the present invention to embed high dimensional data objects may comprise the following steps: (1) generating an initial formal embedding of selected high-dimensional data objects using any of the traditional formal embedding techniques; (2a) designing a deep embedding architecture, which includes choosing the types and numbers of inputs and outputs, types and number of layers, types of units/nonlinearities, and types of pooling, for example, among other design choices, typically in a convolutional neural network; (2b) designing a training strategy; (2c) tuning the parameters of a deep embedding architecture to reproduce, as reliably as possible, the generated embedding for each training sample; (3) optionally deploying the trained deep embedding architecture to convert new high dimensional data objects into approximately the same embedded space as found in step (1); and optionally (4) feeding the computed embeddings of high dimensional objects to an application in a deployed embodiment.