Abstract: Apparatus and method to detect concealed weapons on a person from a distance coordinates movement of a field of view (FOV) of a wide angle view (WAV) camera with movement of an FOV of a millimeter wave (MMW) camera. In an embodiment, the apparatus identifies and tracks a plurality of individuals within the WAV camera FOV. The MMW camera is focused on a first of the identified plurality of individuals, and takes an image of the individual within the MMW camera FOV. In an embodiment, the first individual is centered in the MMW camera FOV. The image is analyzed to detect whether the individual is carrying any concealed weapons. If concealed weapons are detected, the apparatus may signal a need to detain and/or apprehend the individual.

Abstract: Method and apparatus to identify field labels from filled forms using image processing to compare two or a few copies of the same kind of form, possibly filled out differently. Filled forms are processed to provide text strings, one for each line of text on each filled form. The text strings are converted to vectors. Vectors from different filled forms are compared to identify common words which are indicative of field labels. In an embodiment, a histogram may be generated to show frequency of occurrence of characters and words, the histogram values also being indicative of field labels.

Abstract: Method and apparatus to match bounding boxes around text to align forms. The approach is less computationally intensive, and less prone to error than text recognition. For purposes of achieving alignment, information per se is not as important as information location. Information within the bounding boxes is not as critical as is the location of the area which the bounding boxes occupy. Scanning artifacts, missing characters, or noise generally do not affect bounding boxes themselves so much as they do the contents of the bounding boxes. Thus, for purposes of form alignment, the bounding boxes themselves are sufficient. Using bounding boxes also avoids misalignment issues that can result from stray marks on a page, for example, from holes punched in a sheet, or from handwritten notations.

Abstract: Method and apparatus to match bounding boxes around text to align forms. The approach is less computationally intensive, and less prone to error than text recognition. For purposes of achieving alignment, information per se is not as important as information location. Information within the bounding boxes is not as critical as is the location of the area which the bounding boxes occupy. Scanning artifacts, missing characters, or noise generally do not affect bounding boxes themselves so much as they do the contents of the bounding boxes. Thus, for purposes of form alignment, the bounding boxes themselves are sufficient. Using bounding boxes also avoids misalignment issues that can result from stray marks on a page, for example, from holes punched in a sheet, or from handwritten notations.

Abstract: A method, a computer readable medium, and a system for cell annotation are disclosed. The method includes receiving at least one new cell image for cell detection; extracting cell features from the at least one new cell image; comparing the extracted cell features to a matrix of cell features of each class to predict a closest class, wherein the matrix of cell features has been generated from at least initial training data comprising at least one cell image; detecting cell pixels from the extracted cell features of the at least one new cell image using the predicted closest class to generate a likelihood map; extracting individual cells from the at least one cell image by segmenting the individual cells from the likelihood map; and performing a machine annotation on the extracted individual cells from the at least one new cell image to identify cells, non-cell pixels, and/or cell boundaries.

Abstract: An artificial neural network system for image classification, formed of multiple independent individual convolutional neural networks (CNNs), each CNN being configured to process an input image patch to calculate a classification for the center pixel of the patch. The multiple CNNs have different receptive field of views for processing image patches of different sizes centered at the same pixel. A final classification for the center pixel is calculated by combining the classification results from the multiple CNNs. An image patch generator is provided to generate the multiple input image patches of different sizes by cropping them from the original input image. The multiple CNNs have similar configurations, and when training the artificial neural network system, one CNN is trained first, and the learned parameters are transferred to another CNN as initial parameters and the other CNN is further trained. The classification includes three classes, namely background, foreground, and edge.

Abstract: A text recognition method and system involves computing a text matching score between an input text and an output candidate text. The text matching score is computed by evaluating respective N-grams of the input text and the output candidate text. The N-grams are compared in pairs for visual similarity by determining N-gram pair scores, which are used to compute the text matching score. The N-gram pair scores are determined using a set of probabilities of confusion between characters contained in the N-grams. The described approach can address inconsistent results that arise from conventional text similarity quantifiers.

Abstract: Image processing is performed on an input image generated from scanning a filled-in document form. The input image is evaluated against a blank version of various document forms in order to identify the form type of the filled-in document form. The evaluation results in identifying one of the blank document forms as a match to the filled-in document form. Each document form has a set of keywords. The evaluation uses a vector of keyword matches in the filled-in document form. Once a blank document form is identified to be match, the filled-in document form may be categorized according to that document form and/or data extracted from the filled-in document may be stored in association with keywords of that document form.

Abstract: A local connectivity feature transform (LCFT) is applied to binary document images containing text characters, to generate transformed document images which are then input into a bi-directional Long Short Term Memory (LSTM) neural network to perform character/word recognition. The LCFT transformed image is a gray scale image where the pixel values encode local pixel connectivity information of corresponding pixels in the original binary image. The transform is one that provides a unique transform score for every possible shape represented as a 3×3 block. In one example, the transform is computed using a 3×3 weight matrix that combines bit coding with a zigzag pattern to assign weights to each element of the 3×3 block, and by summing up the weights for the non-zero elements of the 3×3 block shape.

Abstract: A 3D scanning system includes a base stand, two circular arc shaped support tracks, a mounting assembly for mounting the support tracks to the base stand with one or more degrees of rotational freedom, two sensor holders mounted on the respective support track for holding two depth sensors, and a drive mechanism for driving the sensor holders to move along the respective support tracks. The mounting assembly supports relative rotation of the two support tracks and pitch and roll rotations of the support tracks. To perform a 3D scan, a stationary object is placed in front of the two depth sensors. The sensor holders are moved along the respective support tracks to different positions to obtain depth images of the objects from different angles, from which a 3D surface of the object is constructed. Prior to scanning, the two depth sensors are calibrated relative to each other.

Abstract: In an optical character recognition (OCR) method for digitizing printed text images using a long-short term memory (LSTM) network, text images are pre-processed using a stroke-aware max-min pooling method before being fed into the network, for both network training and OCR prediction. During training, an average stroke thickness is computed from the training dataset. Stroke-aware max-min pooling is applied to each text line image, where minimum pooling is applied if the stroke thickness of the line is greater than the average stroke thickness, while max pooling is applied if the stroke thickness is less than or equal to the average stroke thickness. The pooled images are used for network training. During prediction, stroke-aware max-min pooling is applied to each input text line image, and the pooled image is fed to the trained LSTM network to perform character recognition.

Abstract: In an optical character recognition (OCR) method for digitizing printed text images using a long-short term memory (LSTM) network, text images are pre-processed using a stroke-aware max-min pooling method before being fed into the network, for both network training and OCR prediction. During training, an average stroke thickness is computed from the training dataset. Stroke-aware max-min pooling is applied to each text line image, where minimum pooling is applied if the stroke thickness of the line is greater than the average stroke thickness, while max pooling is applied if the stroke thickness is less than or equal to the average stroke thickness. The pooled images are used for network training. During prediction, stroke-aware max-min pooling is applied to each input text line image, and the pooled image is fed to the trained LSTM network to perform character recognition.

Abstract: An artificial neural network system for image classification, formed of multiple independent individual convolutional neural networks (CNNs), each CNN being configured to process an input image patch to calculate a classification for the center pixel of the patch. The multiple CNNs have different receptive field of views for processing image patches of different sizes centered at the same pixel. A final classification for the center pixel is calculated by combining the classification results from the multiple CNNs. An image patch generator is provided to generate the multiple input image patches of different sizes by cropping them from the original input image. The multiple CNNs have similar configurations, and when training the artificial neural network system, one CNN is trained first, and the learned parameters are transferred to another CNN as initial parameters and the other CNN is further trained. The classification includes three classes, namely background, foreground, and edge.

Abstract: An artificial neural network system for image classification, including multiple independent individual convolutional neural networks (CNNs) connected in multiple stages, each CNN configured to process an input image to calculate a pixelwise classification. The output of an earlier stage CNN, which is a class score image having identical height and width as its input image and a depth of N representing the probabilities of each pixel of the input image belonging to each of N classes, is input into the next stage CNN as input image. When training the network system, the first stage CNN is trained using first training images and corresponding label data; then second training images are forward propagated by the trained first stage CNN to generate corresponding class score images, which are used along with label data corresponding to the second training images to train the second stage CNN.

Abstract: A 3D scanning system includes a base stand, two circular arc shaped support tracks, a mounting assembly for mounting the support tracks to the base stand with one or more degrees of rotational freedom, two sensor holders mounted on the respective support track for holding two depth sensors, and a drive mechanism for driving the sensor holders to move along the respective support tracks. The mounting assembly supports relative rotation of the two support tracks and pitch and roll rotations of the support tracks. To perform a 3D scan, a stationary object is placed in front of the two depth sensors. The sensor holders are moved along the respective support tracks to different positions to obtain depth images of the objects from different angles, from which a 3D surface of the object is constructed. Prior to scanning, the two depth sensors are calibrated relative to each other.

Abstract: Pathological analysis needs instance-level labeling on a histologic image with high accurate boundaries required. To this end, embodiments of the present invention provide a deep model that employs the DeepLab basis and the multi-layer deconvolution network basis in a unified model. The model is a deeply supervised network that allows to represent multi-scale and multi-level features. It achieved segmentation on the benchmark dataset at a level of accuracy which is significantly beyond all top ranking methods in the 2015 MICCAI Gland Segmentation Challenge. Moreover, the overall performance of the model surpasses the state-of-the-art Deep Multi-channel Neural Networks published most recently, and the model is structurally much simpler, more computational efficient and weight-lighted to learn.

Abstract: A method, a computer readable medium, and a system for cell annotation are disclosed. The method includes receiving at least one new cell image for cell detection; extracting cell features from the at least one new cell image; comparing the extracted cell features to a matrix of cell features of each class to predict a closest class, wherein the matrix of cell features has been generated from at least initial training data comprising at least one cell image; detecting cell pixels from the extracted cell features of the at least one new cell image using the predicted closest class to generate a likelihood map; extracting individual cells from the at least one cell image by segmenting the individual cells from the likelihood map; and performing a machine annotation on the extracted individual cells from the at least one new cell image to identify cells, non-cell pixels, and/or cell boundaries.

Abstract: In a text line segmentation process, connected components (CCs) in document image are categorized into three subsets (normal, large, small) based on their sizes. The centroids of the normal size CCs are used to perform line detection using Hough transform. Among the detected candidate lines, those with line bounding box heights greater than a certain height are removed. For each normal size CC, if its bounding box does not overlap the bounting box of any line with an overlap area greater than a predefined fraction of the CC bounding box, a new line is added for this CC, which passes through the centroid of the CC and has an average slant angle. Each large size CCs are broken into two or more CCs. All CCs are then assigned to the nearest lines. A refinement method is also described, which can take any text line segmentation result and refine it.

Abstract: In a text line segmentation process, connected components (CCs) in document image are categorized into three subsets (normal, large, small) based on their sizes. The centroids of the normal size CCs are used to perform line detection using Hough transform. Among the detected candidate lines, those with line bounding box heights greater than a certain height are removed. For each normal size CC, if its bounding box does not overlap the bounting box of any line with an overlap area greater than a predefined fraction of the CC bounding box, a new line is added for this CC, which passes through the centroid of the CC and has an average slant angle. Each large size CCs are broken into two or more CCs. All CCs are then assigned to the nearest lines. A refinement method is also described, which can take any text line segmentation result and refine it.

Abstract: An artificial neural network system implemented on a computer for cell segmentation and classification of biological images. It includes a deep convolutional neural network as a feature extraction network, a first branch network connected to the feature extraction network to perform cell segmentation, and a second branch network connected to the feature extraction network to perform cell classification using the cell segmentation map generated by the first branch network. The feature extraction network is a modified VGG network where each convolutional layer uses multiple kernels of different sizes. The second branch network takes feature maps from two levels of the feature extraction network, and has multiple fully connected layers to independently process multiple cropped patches of the feature maps, the cropped patches being located at a centered and multiple shifted positions relative to the cell being classified; a voting method is used to determine the final cell classification.