An electronic device and method for human skin detection based on a human body-prior is provided. A color image of a person is acquired, and a 3D body model of the person is estimated based on the color image. One or more unclothed parts of the 3D body model are identified. The one or more unclothed parts correspond to one or more body parts, of which at least a portion of skin remains uncovered by clothes. From the color image, pixel information corresponding to the one or more unclothed parts is extracted and classification information is determined based on the pixel information. The classification information includes a set of values, each indicating a likelihood of whether a corresponding pixel of the color image is part of an unclothed skin portion of the person. The unclothed skin portion is detected in the color image based on the classification information.

Example systems and methods for lesion detection are described herein. An example system includes at least one processor and a memory operably coupled to the at least one processor. The system also includes a candidate selection module configured to receive an image, determine a plurality of candidate points in the image, and select a respective volumetric region centered by each of the candidate points. A portion of a lesion has a high probability of being determined as a candidate point. The system further includes a deep learning network configured to receive the respective volumetric regions selected by the candidate selection module, and determine a respective probability of each respective volumetric region to contain the lesion. Additionally, example methods for training a deep learning network to detect lesions are described herein.

Methods and systems for detecting a vertebral fracture within an x-ray. One method includes receiving a chest x-ray image and identifying a plurality of vertebrae represented in the chest x-ray image. The method further includes extracting a plurality of image patches from the chest x-ray image, each image patch of the plurality of image patches including a portion of the chest x-ray image representing one of the plurality of vertebrae identified in the chest x-ray image. The method further includes sequencing the plurality of image patches into an ordered sequence of image patches, and assigning, with a deep learning model applied to the ordered sequence of image patches, a classification to each of the plurality of image patches indicating whether the image patch represents a fractured vertebra or an unfractured vertebra.

A product defect detection method, device and system are disclosed. The method comprises: constructing a defect detection framework including segmentation networks, a concatenating network and a classification network, and setting a quantity of the segmentation network according to product defect types, wherein each segmentation network corresponds to a defect type; concatenating the sample image with the mask image output by each segmentation network by using the concatenating network to obtain a concatenated image; training the classification network by using the concatenated images to obtain a classification network capable of correctly identifying a product defect and a defect type; and when performing product defect detection, inputting a product image acquired into the defect detection framework, and detecting a product defect and a defect type existing in the product by using the segmentation networks, the concatenating network and the classification network.

A target detection method and apparatus, in which the method includes: obtaining a target candidate region in a to-be-detected image; determining at least two part candidate regions from the target candidate region by using an image segmentation network, where each part candidate region corresponds to one part of a to-be-detected target; and extracting, from the to-be-detected image, local image features corresponding to the part candidate regions; and learning the local image features of the part candidate regions by using a bidirectional long short-term memory LSTM network, to obtain a part relationship feature used to describe a relationship between the part candidate regions; and detecting the to-be-detected target in the to-be-detected image based on the part relationship feature. As a result, image data processing precision in target detection can be improved, application scenarios of target detection can be diversified, and target detection accuracy can be improved.

A method for training a filter-based text recognition system for cataloging image portions associated with files using text from the image portions, the method comprising: receiving a first set of text represented in a first image portion associated with a first file; classifying the first image portion into a predetermined group, wherein the classifying is based at least in part on the first set of text; extracting a first set of features from the first set of text; harmonizing existing data in the predetermined group with the first set of text to modify the first set of features; categorizing the first set of text; and determining analytics-based rules based at least in part on the first set of features.

An image processing system and related method. The system comprises an input interface (IN) configured for receiving an input image. A filter (FIL) of the system filters said input image to obtain a structure image from said input image, said structure image including a range of image values. A range identifier (RID) of the system identifies, based on an image histogram for the structure image, an image value sub-range within said range. The sub-range being associated with a region of interest. The system output through an output interface (OUT) a specification for said image value sub-range. In addition or instead, a mask image for the region of interest or for region or low information is output.

An image recognition method and system based on deep learning are provided. The image recognition system includes a first recognizing engine, at least one second recognizing engine and a processing circuit. The second recognizing engine is activated to recognize a testing image when the first recognizing engine is recognizing the testing image. The processing circuit determines whether to interrupt the first recognizing engine recognizing the testing image according to a result outputted by the second recognizing engine after the second recognizing engine completes recognition of the testing image.

In one aspect, hierarchical image segmentation is applied to an image formed of a plurality of pixels, by classifying the pixels according to a hierarchical classification scheme, in which at least some of those pixels are classified by a parent level classifier in relation to a set of parent classes, each of which is associated with a subset of child classes, and each of those pixels is also classified by at least one child level classifier in relation to one of the subsets of child classes, wherein each of the parent classes corresponds to a category of visible structure, and each of the subset of child classes associated with it corresponds to a different type of visible structure within that category.

Techniques for synthetic training data generation for machine learning are described. A user provides a synthetic training data generator with first set of images and a corresponding set of class identifiers each indicating a type of object depicted in a corresponding image. Each image depicts an object with a substantially or completely monochromatic or transparent background. A user also provides or identifies a second set of images to be used as backgrounds. The synthetic training data generator generates a plurality of images by overlaying one or more of the objects depicted in the first set of images over ones of the second set of images, and further generates labels for each of the plurality of images. The plurality of images and labels are used to automatically train and deploy a machine learning (ML) model.