Abstract: A method for generating a super resolution image may comprise up-scaling an input low resolution image; determining a directivity for each patch included in the up-scaled image; selecting an orientation-specified neural network or an orientation-non-specified neural network according to the directivity of the patch; applying the selected neural network to the patch; and obtaining a super resolution image by combining one or more patches output from the orientation-specified neural network and the orientation-non-specified neural network.

Abstract: The present disclosure provides a method and a device for training a neural network model for use in analyzing captured images, and an intelligent image capturing apparatus employing the same. The neural network model can be trained by performing the image reconstruction and the image classification using based on image data received from a plurality of image capturing devices installed in the monitoring area, calculating at least one loss function based on data processed by the neural network model or the neural network model training device, and determining parameters minimizing the loss function. In addition, the neural network model can be updated through the re-training taking into account the newly acquired image data. Accordingly, the image analysis neural network model can operate with high precision and accuracy.

Abstract: A method for removing compressed Poisson noises in an image, based on deep neural networks, may comprise generating a plurality of block-aggregation images by performing block transform on low-frequency components of an input image; obtaining a plurality of restored block-aggregation images by inputting the plurality of block-aggregation images into a first deep neural network; generating a low-band output image from which noises for the low-frequency components are removed by performing inverse block transform on the plurality of restored block-aggregation images; and generating an output image from which compressed Poisson noises are removed by adding the low-band output image to a high-band output image from which noises for high-frequency components of the input image are removed.

Abstract: The present disclosure provides a method and a device for training a neural network model for use in analyzing captured images, and an intelligent image capturing apparatus employing the same. The neural network model can be trained by performing the image reconstruction and the image classification using based on image data received from a plurality of image capturing devices installed in the monitoring area, calculating at least one loss function based on data processed by the neural network model or the neural network model training device, and determining parameters minimizing the loss function. In addition, the neural network model can be updated through the re-training taking into account the newly acquired image data. Accordingly, the image analysis neural network model can operate with high precision and accuracy.

Abstract: A method for generating a super resolution image may comprise up-scaling an input low resolution image; determining a directivity for each patch included in the up-scaled image; selecting an orientation-specified neural network or an orientation-non-specified neural network according to the directivity of the patch; applying the selected neural network to the patch; and obtaining a super resolution image by combining one or more patches output from the orientation-specified neural network and the orientation-non-specified neural network.

Abstract: A method for removing compressed Poisson noises in an image, based on deep neural networks, may comprise generating a plurality of block-aggregation images by performing block transform on low-frequency components of an input image; obtaining a plurality of restored block-aggregation images by inputting the plurality of block-aggregation images into a first deep neural network; generating a low-band output image from which noises for the low-frequency components are removed by performing inverse block transform on the plurality of restored block-aggregation images; and generating an output image from which compressed Poisson noises are removed by adding the low-band output image to a high-band output image from which noises for high-frequency components of the input image are removed.

Abstract: Provided is a sensory effect adaptation method performed by an adaptation engine, the method including identifying first metadata associated with an object in a virtual world and used to describe the object and converting the identified first metadata into second metadata to be applied to a sensory device in a real world, wherein the second metadata is obtained by converting the first metadata based on a scene determined by a gaze of a user in the virtual world.

Abstract: An apparatus and a method for predicting error possibility, including: generating a first annotation for input data for training by using an algorithm; performing a machine-learning for an annotation evaluation model based on the first annotation and a correction history for the first annotation; generating a second annotation for input data for evaluating by using the algorithm; and predicting the error probability of the second annotation based on the annotation evaluation model are provided.

Abstract: Disclosed is a sensory information providing apparatus. The sensory information providing apparatus may comprise a learning model database storing a plurality of learning models related to sensory effect information with respect to a plurality of videos; and a video analysis engine generating the plurality of learning models by extracting sensory effect association information by analyzing the plurality of videos and sensory effect meta information of the plurality of videos, and extracting sensory information corresponding to an input video stream by analyzing the input video stream based on the plurality of learning model.

Abstract: A method and an apparatus for authoring a machine learning-based immersive media are provided. The apparatus determines an immersive effect type of an original image of image contents to be converted into an immersive media by using an immersive effect classifier learned using an existing immersive media that the immersive effect is already added to an image, detects an immersive effect section of the original image based on the immersive effect type determination result, and generates metadata of the detected immersive effect section.

Abstract: Provided is a sensory effect adaptation method performed by an adaptation engine, the method including identifying first metadata associated with an object in a virtual world and used to describe the object and converting the identified first metadata into second metadata to be applied to a sensory device in a real world, wherein the second metadata is obtained by converting the first metadata based on a scene determined by a gaze of a user in the virtual world.

Abstract: Disclosed is a multi-point connection control apparatus and method for a video conference service. The apparatus may include a front end processor configured to receive video streams and audio streams from user terminals of participants using the video conference service, and generate screen configuration information for providing the video conference service based on the received video streams and the received audio streams, and a back end processor configured to receive at least one of the video streams, at least one of the audio streams, and the screen configuration information from the front end processor, and generate a mixed video for the video conference service based on the received at least one of the video streams, at least one of the audio streams, and the screen configuration information.

Abstract: Provided are a method of providing a video conference service and apparatuses performing the same, the method including determining contributions of a plurality of participants to a video conference based on first video signals and first audio signals of devices of the plurality of participants participating in the video conference, and generating a second video signal and a second audio signal to be transmitted to the devices of the plurality of participants based on the contributions.

Abstract: Disclosed is a sensory information providing apparatus. The sensory information providing apparatus may comprise a learning model database storing a plurality of learning models related to sensory effect information with respect to a plurality of videos; and a video analysis engine generating the plurality of learning models by extracting sensory effect association information by analyzing the plurality of videos and sensory effect meta information of the plurality of videos, and extracting sensory information corresponding to an input video stream by analyzing the input video stream based on the plurality of learning model.

Abstract: A method and an apparatus for authoring a machine learning-based immersive media are provided. The apparatus determines an immersive effect type of an original image of image contents to be converted into an immersive media by using an immersive effect classifier learned using an existing immersive media that the immersive effect is already added to an image, detects an immersive effect section of the original image based on the immersive effect type determination result, and generates metadata of the detected immersive effect section.

Abstract: An apparatus and method capable of recommending content suitable for a user using emotion annotation information is provided. The emotion-based content recommendation apparatus includes a content annotation information database (DB) configured to store basic annotation information and emotion information for each content; a user profile information DB configured to store preferred emotion information in addition to basic profile information for each user; and a content recommendation management module configured to recommend a content list suitable for an emotion of a user based on the emotion information for each content and the preferred emotion information for each user.

Abstract: Disclosed herein are a metaverse client terminal and method for providing a metaverse space capable of enabling interaction between users. The metaverse client terminal includes a sensing data collection unit, a motion state determination unit, a server interface unit, and a metaverse space provision unit. The sensing data collection unit collects sensing data regarding a motion of a first user. The motion state determination unit determines a motion state of the first user, and generates state information data of the first user. The server interface unit transmits the state information data of the first user to a metaverse server, and receives metaverse information data and state information data of a second user. The metaverse space provision unit generates a metaverse space, generates a first avatar and a second avatar, incorporates the first and second avatars into the metaverse space, and provides the metaverse to the first user.

Abstract: Disclosed are a method and an apparatus for generating metadata of immersive media and disclosed also are an apparatus and a method for transmitting metadata related information. The apparatus includes: at least one of a camera module photographing or capturing the image; a gyro module sensing horizontality; a global positioning sensor (GPS) module calculating a position by receiving a satellite signal; and an audio module recording audio; and a network module receiving sensor effect information from a sensor aggregator through a wireless communication network; and an application generating metadata by performing timer-synchronization of an image photographed based on the camera module, a sensor effect collected by using the gyro module or the GPS module, or audio collected based on the audio module.

Abstract: An image encoding method using a Binary Partition Tree (BPT) includes performing the BPT on a reference frame, detecting blocks, each having a difference in a pixel value exceeding a threshold value in a current frame, based on a result of the BPT of the reference frame, and performing the BPT of the current frame on the detected blocks. In accordance with the present invention, block partition is not applied to all frames, but a partial partition method based on a difference between the pixel values of a reference frame and a current frame to be encoded is provided. Accordingly, the encoding speed within the P frame or the B frame can be improved. Furthermore, the PSNR of a corresponding frame can be maintained within a specific range of the PSNR of a reference frame, and a compression effect can be improved.