Rakuten, Inc.

Rakuten Group, Inc.

Jiu Xu

Bjorn Stenger

Tony Tung

Yeongnam Chae

Tomoyuki Mukasa

Bjoern Dietmar Rafael Stenger

Lu Yan

An image processing device according to one embodiment includes an acquisition unit, a generation unit, a calculation unit, and an estimation unit. The acquisition unit acquires an input image. The generation unit generates a plurality of comparison images by superimposing each of a plurality of noises with different densities from each other on a target region being at least part of the input image. The calculation unit calculates, for each of the plurality of comparison images, the degradation level of the comparison image with respect to the input image. The estimation unit estimates the noise level of the input image based on a plurality of calculated degradation levels.

Image processing device, image processing method, and image processing program

To increase accuracy in extracting a foreground area while saving a user time and effort. Image obtaining means of an image processing device obtains an image including a background and an object. Element area setting means sets, with respect to the image, a plurality of element areas respectively corresponding to a plurality of elements in the image. Overlapped area specifying means specifies an overlapped area in which a degree of overlap of the element areas is greater than or equal to a predetermined value in the image. Foreground area extracting means extracts a foreground area corresponding to the object from the image based on the overlapped area.

Image processing device, image processing method, and program

An image processing device according to one embodiment includes a processor. The processor executes a step of acquiring an input image, a step of calculating a feature residual by processing the input image in a convolutional layer, a step of performing at least one convolution on the input image, a step of generating an output feature by applying the feature residual to the convolved input image, and a step of generating an image residual based on the output feature. The image residual is applied to the input image, and thereby a high-resolution image with higher resolution than the input image is generated.

A computer implemented method using artificial intelligence for matching images with locations and directions by acquiring a plurality of panoramic images, detecting objects and their locations in each of the panoramic images, acquiring a floorplan image, detecting objects and their locations in the floorplan image, comparing the objects and locations detect in each of the panoramic image to the objects and locations detected in the floorplan image, and determining a location in the floorplan image where each panoramic image was taken.

Method and apparatus for assigning image location and direction to a floorplan diagram based on artificial intelligence

A 3D model generating system for accurately estimating a shape of an entire space surrounded by a plurality of surfaces is provided. Image acquiring means of the 3D model generating system acquires a plurality of images respectively capturing different directions in a space surrounded by a plurality of surfaces. Partial model generating means generates a plurality of partial models respectively corresponding to the images, each of the partial models representing a portion of the surfaces captured in the images. Overall model generating means generates an overall model representing an entire surface based on the partial models generated by the partial model generating means.

3D model generating system, 3D model generating method, and program

Processing load on a computer can be reduced while saving the labor of photographing an object to estimate a size of the object. Photographed image acquiring means of a 3D model generating system acquires a photographed image in which a first object and a second object are disposed in a real space. Virtual image generating means generates a virtual image in which a first model representing the first object that is a standard size and a second model representing the second object that is not the standard size are disposed in a virtual space. Parameter determining means determines a size parameter indicating a size of the second object based on a result of comparing the photographed image with the virtual image.

A method of using a computer to semantically segment an image using a convolutional neural network system where a processor configured to convolve an input image with a plurality of filters and outputting a first output volume, pool the first output volume and creating a first activation map, determine the level of influence of the first activation map on the semantic segmentation, up-pool the first activation map to form an output image having a same number of pixels as the input image, output a probabilistic segmentation result, labeling each pixel's probability that it is a particular label, and the determination of the level of influence of the first activation map on the semantic segmentation is done using a gate layer that is positioned between a pooling layer and an up-pooling layer.

Skip architecture neural network machine and method for improved semantic segmentation

A system for and method for improved computer vision image processing and image segmentation and recognition is disclosed. The system and method utilizes a cost function approach for improved image segmentation and recognition. In particular, a global cost function is defined and then the global cost function is minimized. This global cost function takes into account two processing pipelines of data determined by the operation of two different processing technologies upon the target input image. Constraints are utilized to ensure data consistency across the model and between the data pipelines. The system and method are useful for processing low quality images containing alphanumeric data such as floorplan images.

Growing community of inventors

Tokyo, Japan

Jiu Xu