Abstract: A knowledge distillation based semantic image segmentation method includes inputting an input image to a teacher network and a student network; normalizing a first feature vector corresponding to each pixel in a feature map of a last layer of the teacher network and normalizing a second feature vector corresponding to each pixel in a feature map of a last layer of the student network; generating the first channel and space association matrix and the second channel and space association matrix based on the normalized first feature vector and the normalized second feature vector, and defining a first loss function based on a Euclidean norm value of the difference between the first channel and space association matrix and the second channel and space association matrix.

Abstract: Disclosed is a stereo matching method of images performed by a system implemented by a computer, including the steps of: receiving a pair of images obtained at different time points; generating a feature map by extracting features of each pixel of the pair of images; generating sequentially cost volumes for partial time points based on the feature map and generating a feature map for the entire disparity by fusing the cost volumes using 2D convolution; and generating a final disparity map by refining the generated feature map.

Abstract: Disclosed is a stereo matching method of images performed by a system implemented by a computer, including the steps of: receiving a pair of images obtained at different time points; generating a feature map by extracting features of each pixel of the pair of images; generating sequentially cost volumes for partial time points based on the feature map and generating a feature map for the entire disparity by fusing the cost volumes using 2D convolution; and generating a final disparity map by refining the generated feature map.

Abstract: Disclosed are a method and an apparatus for processing a blurred image. The method for processing a blurred image includes the steps of generating a first input feature map and a second input feature map with a feature distribution for blur removal from the blurred image, generating a prediction feature map from the first input feature map by using a self-spatial feature transform (SSFT) module which transforms the feature distribution for blur removal into a feature distribution for face recognition without external information, and generating a deblurred image based on the second input feature map and the prediction feature map.

Abstract: The present disclosure provides a knowledge distillation based semantic image segmentation method including inputting an input image to a teacher network and a student network; normalizing a first feature vector corresponding to each pixel in a feature map of a last layer of the teacher network and normalizing a second feature vector corresponding to each pixel in a feature map of a last layer of the student network; generating the first channel and space association matrix and the second channel and space association matrix based on the normalized first feature vector and the normalized second feature vector, and defining a first loss function based on an Euclidean norm value of the difference between the first channel and space association matrix and the second channel and space association matrix.

Abstract: A multiview image display apparatus is provided. The multiview image display apparatus includes a depth adjuster configured to adjust a depth of an input image, a renderer configured to perform rendering on multiview images based on the depth-adjusted image, a crosstalk compensator configured to perform crosstalk inverse compensation on the rendered multiview images, a display configured to arrange the crosstalk inverse compensated multiview images in a preset arrangement pattern and display the crosstalk inverse compensated multiview images arranged in the preset arrangement pattern, and a controller configured to estimate crosstalk, and control the depth adjuster to adjust a depth value of a region that satisfies a preset condition to a preset depth value based on the estimated crosstalk.

Abstract: A video processing apparatus is provided. The video processing apparatus includes: a storage configured to store a video captured by a camera; and a controller configured to separate a target object from a background image in each of a plurality of key frames contained in the video, stitch separated background images to generate a panoramic background image, synthesize the target object with the panoramic background image to acquire a plurality of panoramic images corresponding to the plurality of key frames, and combine the panoramic images in a time axis to generate a panoramic video.

Abstract: A video processing apparatus is provided. The video processing apparatus includes: a storage configured to store a video captured by a camera; and a controller configured to separate a target object from a background image in each of a plurality of key frames contained in the video, stitch separated background images to generate a panoramic background image, synthesize the target object with the panoramic background image to acquire a plurality of panoramic images corresponding to the plurality of key frames, and combine the panoramic images in a time axis to generate a panoramic video.

Abstract: A multiview image display apparatus is provided. The multiview image display apparatus includes a depth adjuster configured to adjust a depth of an input image, a renderer configured to perform rendering on multiview images based on the depth-adjusted image, a crosstalk compensator configured to perform crosstalk inverse compensation on the rendered multiview images, a display configured to arrange the crosstalk inverse compensated multiview images in a preset arrangement pattern and display the crosstalk inverse compensated multiview images arranged in the preset arrangement pattern, and a controller configured to estimate crosstalk, and control the depth adjuster to adjust a depth value of a region that satisfies a preset condition to a preset depth value based on the estimated crosstalk.

Abstract: A method of processing frames obtained by multiple exposures may include categorizing, by at least one processor, a plurality of frames captured by multiple exposures into one of a reference frame and a target frame, compensating for global motion resulting from motion of a capturing apparatus capturing an object, and compensating for local motion resulting from motion of the object, based on a joint histogram between the reference frame and the target frame.

Abstract: An apparatus and method creating a ghost-free High Dynamic Range Image (HDRI) based on filtering are provided. It is possible to effectively prevent a ghost phenomenon from occurring when a single HDRI is created from a plurality of LDRIs, by defining a ghost area using a probability based on a global transfer function indicating a relationship for intensities of several frames, rather than searching for or identifying a ghost area in a single or each image.

Abstract: An apparatus and method creating a ghost-free High Dynamic Range Image (HDRI) based on filtering are provided. It is possible to effectively prevent a ghost phenomenon from occurring when a single HDRI is created from a plurality of LDRIs, by defining a ghost area using a probability based on a global transfer function indicating a relationship for intensities of several frames, rather than searching for or identifying a ghost area in a single or each image.

Abstract: A method of processing frames obtained by multiple exposures may include categorizing, by at least one processor, a plurality of frames captured by multiple exposures into one of a reference frame and a target frame, compensating for global motion resulting from motion of a capturing apparatus capturing an object, and compensating for local motion resulting from motion of the object, based on a joint histogram between the reference frame and the target frame.

Abstract: A cyclone dust collector and a vacuum cleaner having the same. The cyclone dust collector has a cyclone body shaped to have a relatively wider upper portion and a relatively narrower lower portion and also has a suction port and a discharge port, a grill member connected to the discharge port, a dust receptacle connected to the cyclone body, and a blocking member for partially blocking the grill member. Air drawn in through the suction port moves in an increasingly wider radial path as it travels upwardly to the discharge port, increasing the centrifugal force on suspended particles as they travel toward the discharge port. The shape of the cyclone body prevents turbulent flow from being generated. Dust is prevented by the blocking member from attaching to the grill member before being centrifuged, performance of the grill member is enhanced.

Abstract: The present invention relates to a compound with aryl-amidines, particularly amidinoaryl-cyclopropanes, amidinoarylmethyl-pyrroles, amidinoaryl-benzenes, amidinoaryl-pyridines, or amindonoaryl-alanines, represented by formula (1), a pharmaceutically acceptable salt, a prodrug, a hydrate, a solvate or an isomer thereof, which are inhibitors of coagulation enzyme, factor Xa (FXa). The present invention also relates to a pharmaceutical composition containing the compound, and a method of using the same as an anticoagulant agent for treatment and prevention of thrombosis disorders.