Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for a computing system that receives images that each have a predefined exposure attribute. For each image, a first set of features of the image are extracted. The first set of features are associated with a label indicating no modification of the image is required. A luminosity characteristic of the image is adjusted to form an adjusted image. A second set of features of the adjusted image are extracted. A neural network is trained to adjust luminosity characteristics of images using the first set of features and the second set of features of the adjusted image. An exposure adjustment model adjusts luminosity characteristics of images based on correction values determined using the trained neural network.

Abstract: The present disclosure relates to a handwritten signature authentication system and method. More particularly, the present disclosure provides a system and method of authenticating a handwritten signature based on dynamic movement tracking of spatial-division segments, in which handwritten signature authentication is performed by handwritten signature characteristics information (i.e.

Abstract: According to one aspect, scene classification may be provided. An image capture device may capture a series of image frames of an environment from a moving vehicle. A temporal classifier may classify image frames with temporal predictions and generate a series of image frames associated with respective temporal predictions based on a scene classification model. The temporal classifier may perform classification of image frames based on a convolutional neural network (CNN), a long short-term memory (LSTM) network, and a fully connected layer. The scene classifier may classify image frames based on a CNN, global average pooling, and a fully connected layer and generate an associated scene prediction based on the scene classification model and respective temporal predictions. A controller of a vehicle may activate or deactivate vehicle sensors or vehicle systems of the vehicle based on the scene prediction.

Abstract: The present disclosure provides a dense crowd counting method and an apparatus, including: acquiring an image to be detected, where the image to be detected includes images of people; feeding the image to be detected into a convolutional neural network model to obtain a crowd density map of the image to be detected; and determining the number of the images of people in the image to be detected according to the crowd density map. Feature information of an image to be detected may be fully extracted through the above mentioned process, thereby realizing a better performance of crowd counting and density estimation, providing great convenience for subsequent security monitoring, crowd control and other applications.

Abstract: The present disclosure relates to an information processing method and an information processing apparatus. The information processing method according to the present disclosure performs training on a classification model by using a plurality of training samples, and comprises the steps of: adjusting a distribution of feature vectors of the plurality of training samples in a feature space based on a typical sample in the plurality of training samples; and performing training on the classification model by using the adjusted feature vectors of the plurality of training samples. Through the technology according to the present disclosure, it is possible to perform pre-adjustment on training samples before training, such that it is possible to reduce discrimination between training samples belonging to a same class and increase discrimination between training samples belonging to different classes in the training process.

Abstract: There is provided a program generating apparatus including a generating unit and a genetic processing unit. The generating unit is configured to generate tree structures each representing an image classification program. Each of the tree structures has a first level group and a second level group. Elements of nodes in the first level group are selected from amongst image filters each used to apply preprocessing to an input image. An element of a node in the second level group is selected from amongst setting programs each used to set a different value as a control parameter for generating a classifier based on information obtained by execution of the elements selected for the nodes in the first level group. The genetic processing unit is configured to output, using genetic programming, a tree structure with a fitness score exceeding a predetermined threshold based on the tree structures.

Abstract: In an aspect of the present disclosure relates to a network involving humans and AI based systems working in conjunction to perform tasks such as traffic violation detection, infrastructure monitoring, traffic flow management, crop monitoring etc. from visual data acquired from numerous data acquisition sources. The system includes a network of electronic mobile devices with AI capabilities, connected to a decentralized network working towards capturing high quality data for finding events or objects of interest in the real world, retraining AI models, for processing the high volumes of data on decentralized or centralized processing units and also being used for the verification of the outputs of the AI models. The system talks about many annotation techniques on smartphones for crowdsourced AI Data Labeling for AI Training.

Abstract: A classifier receives a document from a multi-document transaction. The classifier analyzes the document to identify one or more embedded dates in the content of the document and context of one or more positions of the one or more embedded dates in the document. The classifier evaluates each of the one or more embedded dates based on the separate context of each of the one or more positions within the document and a relative age of the one or more embedded dates in view of temporal characteristics of multiple categories of documents of a transaction to select a particular category associated with the document from among the multiple categories. The classifier classifies the document within the transaction as a particular logical type identified by the particular category from among multiple logical types.

Abstract: The traffic signal recognition device includes a traffic signal recognition unit configured to perform processing for recognizing the traffic signal based on the result of imaging performed by a camera, an external situation recognition unit configured to recognize a size and position of a surrounding vehicle, and a occluded situation determination unit configured to determine whether or not the area in front of a host vehicle is in the traffic signal occluded situation, in which the line of sight from the camera to the traffic signal is blocked by the surrounding vehicle. The traffic signal recognition unit is configured not to perform the processing for recognizing the traffic signal within a difficulty zone and not to perform the processing for recognizing the traffic signal if it is determined that the area in front of the host vehicle is in the traffic signal occluded situation.

Abstract: An angularly-dependent reflectance of a surface of an object is measured. Images are collected by a sensor at different sensor geometries and different light-source geometries. A point cloud is generated. The point cloud includes a location of a point, spectral band intensity values for the point, an azimuth and an elevation of the sensor, and an azimuth and an elevation of a light source. Raw pixel intensities of the object and surroundings of the object are converted to a surface reflectance of the object using specular array calibration (SPARC) targets. A three-dimensional (3D) location of each point in the point cloud is projected back to each image using metadata from the plurality of images, and spectral band values are assigned to each value in the point cloud, thereby resulting in a multi-angle spectral reflectance data set.

Abstract: A classifier receives a document and analyzes the document to determine one or more predicted roles of one or more signatories, each predicted role determined based on one or more signature elements in the content of the document executed by the one or more signatories. The classifier evaluates each of the one or more predicted roles in view of a plurality of expected signatory role characteristics of a plurality of categories of documents of a transaction to select a particular category associated with the document from among the plurality of categories. The classifier classifies the document within the transaction as a particular logical type identified by the particular category from among a plurality of logical types for the transaction.

Abstract: A method for deriving an image sensor's 3D pose estimate from a 2D scene image input includes at least one Machine Learning algorithm trained a priori to generate a 3D depth map estimate from the 2D image input, which is used in conjunction with physical attributes of the source imaging device to make an accurate estimate of the imaging device 3D location and orientation relative to the 3D content of the imaged scene. The system may optionally employ additional Machine Learning algorithms to recognize objects within the scene to further infer contextual information about the scene, such as the image sensor pose estimate relative to the floor plane or the gravity vector. The resultant refined imaging device localization data can be applied to static (picture) or dynamic (video), 2D or 3D images, and is useful in many applications, most specifically for the purposes of improving the realism and accuracy of primarily static, but also dynamic Augmented Reality (AR) applications.

Abstract: One or more neural networks generate a first vector field from an input image and a reference image. The first vector field is applied to the input image to generate a first warped image. The training of the neural networks is evaluated via one or more objective functions. The neural networks are updated in response to the evaluating. The neural networks generate a second vector field from the input image and the reference image. A number of degrees of freedom in the first vector field is less than a number of degrees of freedom in the second vector field. The second vector field is applied to the input image to generate a second warped image. The neural networks are evaluated via the one or more objective functions, the reference image and the second warped image. The networks are updated in response to the evaluating.

Abstract: In a method and apparatus for training a computer system for use in classification of an image by processing image data representing the image, image data are compressed and then loaded into a programmable quantum annealing device that includes a Restricted Boltzmann Machine. The Restricted Boltzmann Machine is trained to act as a classifier of image data, thereby providing a trained Restricted Boltzmann Machine; and, the trained Restricted Boltzmann Machine is used to initialize a neural network for image classification thereby providing a trained computer system for use in classification of an image.

Abstract: According to one implementation, a pose estimation and body tracking system includes a computing platform having a hardware processor and a system memory storing a software code including a tracking module trained to track motions. The software code receives a series of images of motion by a subject, and for each image, uses the tracking module to determine locations corresponding respectively to two-dimensional (2D) skeletal landmarks of the subject based on constraints imposed by features of a hierarchical skeleton model intersecting at each 2D skeletal landmark. The software code further uses the tracking module to infer joint angles of the subject based on the locations and determine a three-dimensional (3D) pose of the subject based on the locations and the joint angles, resulting in a series of 3D poses. The software code outputs a tracking image corresponding to the motion by the subject based on the series of 3D poses.

Abstract: Techniques are disclosed for document boundary detection (BD) from an input image using a combination of deep learning model and image processing algorithms. Quadrilaterals approximating the document boundaries in the input image are determined and rated separately using both these approaches: deep leaning using convolutional neural network (CNN) and heuristics using image processing algorithms. Thereafter, the best rated quadrilateral is selected from the quadrilaterals obtained from both the approaches.

Abstract: Introduced here is a technique to create small compressed image files while preserving data quality upon decompression. Upon receiving an uncompressed data, such as an image, a video, an audio, and/or a structured data, a machine learning model identifies an object in the uncompressed data such as a house, a dog, a text, a distinct audio signal, a unique data pattern, etc. The identified object is compressed using a compression treatment optimized for the identified object. The identified object, either before or after the compression, is removed from the uncompressed data. The uncompressed data with the identified object removed is compressed using a standard compression treatment.

Abstract: A system and method for wafer-level inspection using on-valve inspection detectors to detect defects on a semiconductor wafer surfaces during a semiconductor device manufacturing process is disclosed herein. In some exemplary embodiments, a method for wafer-level inspection includes: transporting a semiconductor wafer through a transfer port of a processing chamber; scanning a surface of the semiconductor wafer automatically using at least one on-valve inspection detector arranged on a vacuum valve providing access through the transfer port; generating at least one surface image of the surface of the semiconductor wafer; and analyzing the at least one surface image to detect defects on the surface of the semiconductor wafer.

Abstract: A method for ascertaining a piece of topological information of an intersection, including locating a vehicle with lane accuracy when negotiating the intersection; ascertaining data by the vehicle when negotiating the intersection; transmitting the data to a processing unit; and ascertaining a connectivity of lane-roadway combinations of the intersection from the data with the aid of the processing unit.

Abstract: Various systems and methods for generating environmentally contextualized patterns are disclosed. The system and method generates the environmentally contextualized pattern from a set of images representing an environment. The color palette of the dominant colors from the representational images is processed to remove the gray hues, set the remaining highest and lowest value hues to a particular contrast, and then determine a split complement from the lowest value hue. An algorithm, such as a reaction-diffusion algorithm, is then utilized to generate a pattern incorporating the aforementioned hues. The pattern generated by the algorithm provides a high degree of visual contrast with the environment that the images represent, allowing an individual wearing the pattern to be readily visually identifiable against the environment.