Chung Ang University Industry Academic Cooperation

Samsung Electronics Co., Ltd.

Korea Electronics Technology Institute

Hyundai Motor Company

Skhynix Inc.

Kia Motors Corporation

Joon Ki Paik

Byeong Ho Choi

Yoon Jong Yoo

Jin Beum Jang

Jin Su Park

Yong Hwan Kim

Seung Won Lee

Chul Ho Lee

Jin Ho Park

Eun Sung Lee

Jae Hoon Jung

Dong Gyun Kim

Chul-Ho Lee

Sang Woo Park

Min Jae Kim

Je Woo Kim

Yun Sup Ann

Tae Beom Lim

Jae Hwan Jeon

Hyun Sang Kim

Yong-Hwan Kim

Jong Heon Kim

Jae Hyun Im

Hyok Song

Wan Hee Jo

Young Chul Youn

Sang-ku Kim

Dong Geun Kim

Sung-Il Su

Jin Sol Choi

Sun Kyu Kwon

Yeong Heon Mok

Yeong Joon Kim

Joong Chol Shin

Heun Seung Lim

Yeon Seung Choo

Dae Hee Kim

In Hye Yoon

Eun Jung Chae

Hee Gwang Kim

Nak Eun Choi

Yoon-Jong Yoo

Joo Young Yi

He Jin Cheong

Ha Sil Park

Disclosed are a deep learning model optimization method and apparatus for medical image segmentation. A deep learning model optimization method for medical image segmentation includes: (a) initializing a model parameter; (b) updating the model parameter by performing model-agnostic meta learning (MAML) on a model based on sample batch and applying a gradient descent algorithm to a loss function; (c) setting an optimizer parameter as the updated model parameter, performing one-shot meta-learning on the model, and then updating the optimizer parameter by applying the gradient descent algorithm to the loss function; and (d) updating the model parameter by reflecting the updated optimizer parameter.

Deep learning model optimization method and apparatus for medical image segmentation

Disclosed are a method and an apparatus for classifying patients with brain disorders based on EEG analysis. The method for classifying patients with brain disorders based on EEG analysis includes: (a) receiving an EEG data set with a clinical diagnosis label, wherein the EEG data set includes a plurality of EEG signals and age information; (b) augmenting the EEG data set through a deep learning-based screening model, and augmenting the EEG signals and the age information in different schemes; (c) training the deep learning-based screening model to classify EEG data of patients into a target clinical diagnosis label by using the augmented EEG signals and age information; and (d) predicting a brain diagnosis label of brain disorders by applying EEG data of patients to the trained deep learning-based screening model.

Method and apparatus for classifying patients with brain disorders based on electro encephalo graphy (EEG) analysis

Disclosed are a training data generation method and apparatus for an artificial intelligence visual analysis model. The training data generation method includes: (a) receiving a background image and object-related information; (b) generating a target object image that meets a condition based on the object-related information; (c) applying the background image to a depth estimation model to estimate depths of each pixel location and generate a depth map; (d) determining an insertion location and scale at which the target object image is inserted based on the depth map to generate a local object insertion image and a mask image, respectively; (e) simply synthesizing the local object insertion image and the background image and then generating a synthesis image reflecting a scale feature of the background image through a trained multi-scale visual analysis model; and (f) generating a final training image using the background image, the synthesis image, and the mask image.

Training data generation method and apparatus for artificial intelligence visual analysis model

Disclosed are a high dynamic range imaging method and apparatus. The high dynamic range imaging method includes: (a) acquiring a first low dynamic range (LDR) image, a second LDR image, and a third LDR image having different exposure values; (b) applying the first, second, and third LDR images to a weight map estimation model to extract a weight map, and integrating the first, second, and third LDR images using the extracted weight map to generate an input feature map; and (c) applying the input feature map to a Swin-Fourier convolution network model to perform Fourier transform and convolution on the input feature map, and then connecting a reference feature map to generate a high dynamic range (HDR) image.

High dynamic range imaging method and apparatus

Disclosed are a method and apparatus for integrating image enhancement and training data generation using a generative recurrent neural network model. The method of integrating image enhancement and training data generation using a generative recurrent neural network model includes: (a) receiving a target image as input, and (b) applying the target image to a trained generative recurrent neural network model to selectively generate any one of a high-quality image with enhanced image quality and a low-quality image depending on a type of the target image.

Method and apparatus for integrating image enhancement and training data generation using a generative recurrent neural network model

Disclosed are multiple object detection method and apparatus. The multiple object detection apparatus includes a feature map extraction unit for extracting a plurality of multi-scale feature maps based on an input image, and a feature map fusion unit for generating a multi-scale fusion feature map including context information by fusing adjacent multi-scale feature maps among the plurality of multi-scale feature maps generated by the feature map extraction unit.

Multiple object detection method and apparatus

An apparatus for detecting an object of a vehicle is provided. The apparatus includes a camera that acquires an image from front of a vehicle and a controller that generates feature pyramid images based on a plurality of feature images extracted from the image. The controller also generates feature aggregation images by filtering the feature pyramid images, detects a pedestrian area from the feature aggregation images, and detects face regions from the feature pyramid images. At least one of the face regions that overlaps the pedestrian area is then determined as a face of a pedestrian.

Apparatus and method for detecting object of a vehicle

Disclosed a normalized metadata generation device, and object occlusion detection device and method. A normalized metadata generation method includes generating a multi-ellipsoid based three-dimensional human model using perspective features of a plurality of two-dimensional images obtained by the multiple cameras, performing scene calibration based on the three-dimensional human model to normalize object information of the object included in the two-dimensional images, and generating normalized metadata of the object from the two-dimensional images on which the scene calibration is performed.

Normalized metadata generation device, object occlusion detection device and method

The apparatus for removing haze from an image using a fuzzy membership function includes memory configured to store computer-readable instructions; and a processor configured to execute the instructions, wherein the processor calculates a hazy value and a transmission map using an input image including a hazy component and generates a restored image from which the hazy component has been removed using the hazy value and the transmission map, and wherein the processor selects a first area in the input image, calculates a fuzzy membership function of pixels of the input image with respect to the hazy value using brightness values of the pixels of the input image and brightness values of pixels of the first area, and calculates the transmission map using the brightness values of pixels of the first area and the fuzzy membership function.

Apparatus and method for removing haze from image using fuzzy membership function, and computer program for performing the method

The present invention relates to an auto-focus system and method based on a sensor, using hierarchical extraction and motion estimation for an image. The auto-focus method may include the steps of: (a) extracting a phase difference image from a sensed image; (b) extracting a robust feature image which satisfies a preset reference with respect to noise and out-of-focus blur, from the phase difference image; and (C) estimating a motion vector for lens shift from the feature image.

Auto-focus system for a digital imaging device and method

A method and an apparatus for restoring a motion blurred image are disclosed. The method of restoring a motion blurred image comprises calculating a motion trajectory and a project trajectory of a camera using an embedded sensor; estimating a point spread function according to camera motion by using the motion trajectory and the project trajectory; and restoring the motion blurred image using the estimated point spread function and a spatially variant activity map.

Method and apparatus for restoring motion blurred image

Provided are a non-dyadic lens distortion correction method and an apparatus thereof. The non-dyadic lens distortion correction method includes: (a) acquiring a non-dyadic lens image photographed through a non-dyadic lens; (b) correcting geometric distortion for the non-dyadic lens image by stage by applying a magnification ratio based on a projection model; and (c) compensating a high-frequency component of an image corrected in a previous step by the patch unit in the non-dyadic lens image of which the geometric distortion is corrected.

Non-dyadic lens distortion correction method and apparatus

Provided are an apparatus and a method for encoding a high fidelity video, and an apparatus and a method for decoding a high fidelity video. A video encoding method for encoding first to third planes having different color components includes performing intra prediction in the first plane or inter prediction between each frame for a block of the first plane, and generating a first residue corresponding to a difference between the block of the first plane and the predicted block of the first plane, predicting a block of the second plane using a reconstructed pixel peripheral to the block of the second plane and a reconstructed pixel peripheral to the block of the first plane corresponding to the reconstructed pixel peripheral to the block of the second plane, and generating a second residue corresponding to a difference between the block of the second plane and the predicted block of the second plane, predicting a block of the third plane using a reconstructed pixel peripheral to the block of the third plane and any one of a reconstructed pixel peripheral to the block of the first plane corresponding to the reconstructed pixel peripheral to the block of the third plane and a reconstructed pixel peripheral to the block of the second plane corresponding to the reconstructed pixel peripheral to the block of the third plane, and generating a third residue corresponding to a difference between the block of the third plane and the predicted block of the third plane, and encoding the first to third residues.

Apparatus and method for encoding and decoding high fidelity video, and computer-readable storage medium

An effective H.264/AVC CAVLC decoding method is disclosed. In accordance with the method a VLC code is classified into groups according to a correlation thereof, an arithmetic equation is defined for each of the groups, and a decoding is carried out according to the arithmetic equation in order to minimize a memory access by a table look-up and reduce a decoding time and a power consumption.

Effective decoding method of H.264/AVC context-based adaptive variable length coding

A block-based image processing method and an apparatus therefor, in which blocking artifacts caused by a block-based image processing can be removed by using spatial adaptive filtering based on image restoration theory. The block-based image processing method, for removing blocking artifacts induced in processing an image in blocks of a predetermined unit size, includes the steps of: (a) dividing the processed image signal into blocks of the predetermined unit size; (b) detecting an edge direction of the image for each block divided; (c) band-pass filtering the processed image signal to obtain plurality of filtered image signals equal in number to the possible types of edge direction; (d) selecting one of the band-pass filtered image signals for each block, according to the detected edge direction; and (e) synthesizing the selected image signals for all the blocks to obtain an original signal from which the blocking artifact is removed.

Block-based image processing method and apparatus therefor

Growing community of inventors

Seoul, South Korea

Joon Ki Paik