Methods and apparatuses for video transcoding based on spatial or temporal importance include: in response to receiving an encoded video bitstream, decoding a picture from the encoded video bitstream; determining a first level of spatial importance for a first region of a background of the picture based on an image segmentation technique; applying to the first region a first resolution-enhancement technique associated with the first level of spatial importance for increasing resolution of the first region by a scaling factor, wherein the first resolution-enhancement technique is selected from a set of resolution-enhancement techniques having different computational complexity levels; and encoding the first region using a video coding standard.

A method of image retargeting is provided. The method includes obtaining a source image, obtaining a target size for a retargeted image based on the source image, generating a two-dimensional importance map for the source image, generating, based on the two-dimensional importance map and the target size, a warping mesh having a distortion metric below a threshold value, determining whether a size of the warping mesh corresponds to the target size, and based on the size of the warping mesh being determined to correspond to the target size, rendering the retargeted image by applying the warping mesh to the source image.

An image processing apparatus includes: a memory; and a processor configured to execute one or more instructions stored in the memory to: determine a scale weight for each pixel of original pixels in an original image based on first attribute-related information set by dividing the original image into an object region and a peripheral region, respectively; determine an additional weight for at least one pixel of the original pixels in the peripheral region, based on the first attribute-related information, and second attribute-related information that is based on an amount of a change between pixel values of adjacent pixels; and obtain a transformed image of which a size is changed from the original image, by applying at least one of the scale weight and the additional weight to corresponding pixels of the original pixels to obtain a pixel value of a transformed pixel in the transformed image.

A visual perception of an arbitrary transparent material is imparted to an arbitrary image. In accordance with each element of each deformation map included in a sequence of deformation maps that correspond to a time series, each element of a target image is moved to obtain each deformed image of the time series. Each element of each of the deformation maps indicates a movement direction and a movement amount of each pixel of the target image corresponding to the element. Each deformation map included in a sequence of deformation maps corresponding to a first time interval in the time series corresponds to each of two-dimensional arrays obtained by moving, in a first direction, elements of two-dimensional arrays corresponding to immediately-previous deformation maps, and each deformation map included in a sequence of deformation maps corresponding to a second time interval in the time series corresponds to each of two-dimensional arrays obtained by moving, in a second direction, elements of two-dimensional arrays corresponding to immediately-previous deformation maps. Here, the first direction and the second direction differ from one another.

Systems and method directed to an inversion-consistent transfer learning framework for generating portrait stylization using only limited exemplars. In examples, an input image is received and encoded using a variational autoencoder to generate a latent vector. The latent vector may be provided to a generative adversarial network (GAN) generator to generate a stylized image. In examples, the variational autoencoder is trained using a plurality of images while keeping the weights of a pre-trained GAN generator fixed, where the pre-trained GAN generator acts as a decoder for the encoder. In other examples, a multi-path attribute aware generator is trained using a plurality of exemplar images and learning transfer using the pre-trained GAN generator.

Disclosed herein is a machine learning application for automatically reading filled-in forms. There are multiple steps involved in using a computer to accurately read a handwritten form. First, the system identifies the form. Second, the system identifies what parts of the form are important. Third, the important parts are extracted as image data (known as shreds). Finally, fourth, the system interprets the shreds. This application is focused on steps two and three of that overall process. The disclosed techniques relate to training a machine learning system on a given series of forms such that when provided future filled-in forms within that series, the system is able to extract the portions of the filled-in form that are important/relevant.

Examples are disclosed relating to applying an analytical geometric projection that has been modified by an amplitude function. One example provides a computing device comprising a logic subsystem and a storage subsystem holding instructions executable by the logic subsystem to receive an image of a scene as acquired by an image sensor, apply a mapping to the image of the scene that maps pixels of the image to projected pixels on an analytical projection that is modified by an amplitude function such that the analytical projection achieves a higher zoom effect on pixels closer to a center of the image compared to pixels closer to an edge of the image, thereby obtaining a corrected image, and output the corrected image.

Methods and apparatus relating to techniques for provision of low power foveated rendering to save power on GPU (Graphics Processing Unit) and/or display are described. In various embodiment, brightness/contrast, color intensity, and/or compression ratio applied to pixels in a fovea region are different than those applied in regions surrounding the fovea region. Other embodiments are also disclosed and claimed.

Systems and methods for an intelligent camera system are provided. A method includes receiving, from a first camera in a vehicle, view data corresponding to an area from a vantage point of the vehicle. The method further includes detecting a region of interest from the view data provided by the first camera. The method also includes providing the region of interest to a second camera in the vehicle. The method further includes receiving, from the second camera, zoom view data corresponding to a zoom view of the region of interest.

A method and apparatus for method for image processing, and a computer storage medium are provided. The method includes: obtaining a first image, identifying a target object in the first image, and obtaining leg detection information of the target object; respectively determining a contour line and a target line of a leg region of the target object based on the leg detection information; and performing image deformation processing on the leg region based on the contour line and the target line of the leg region to generate a second image.