Realtek Semiconductor Inc.

Yi-Ting Bao

Yen-Lin Chen

Shang-Yen Lin

Chia-Wei Yu

Chia-wei Yu

Cheng-yu Kuan

Hao-Ran Wang

A super resolution (SR) image generation circuit includes an image scale-up circuit, a stable SR processing circuit, a generative adversarial network (GAN) processing circuit, and a configurable basic block pool circuit. The image scale-up circuit is arranged to receive and process an input image to generate a scaled-up image. The stable SR processing circuit is arranged to receive a feature map of the input image to generate a stable delta value. The GAN processing circuit is arranged to receive the feature map to generate a GAN delta value. The configurable basic block pool circuit is arranged to dynamically configure a plurality of basic blocks according to a depth requirement of the input image, to generate a configuration result. The SR image generation circuit generates an SR image according to the scaled-up image, the stable delta value, and the GAN delta value.

Super resolution image generation circuit

The present invention discloses an image enlarging method having super resolution enlarging mechanism that includes the steps outlined below. An enlarging module of a neural network system receives an input image to generate an enlarged image. A front end convolutional path included in a neural network module of the neural network system receives the input image to perform convolution to generate a front end operation output result. Branching convolutional paths included in the neural network module respectively receive the front end operation output result to perform convolution to generate groups of output image residues. A mixing module of the neural network system weights the output image residues according to weighing settings related to image regions of the input image and mixes the weighted output image residues to generate a group of final output image residue such that an enhancement module enhances the enlarged image to generate an output enlarged image.

Image enlarging apparatus and method having super resolution enlarging mechanism

A system for image scaling and enhancement is provided. The system includes a scaling processing unit, a deep-learning residue network unit and a combination unit. The filter scaling processing unit is configured to upscale a low-resolution image to output a high-resolution image. The deep-learning residue network unit is operated based on a deep-learning result, and configured to output a high-resolution residue image corresponding to the low-resolution image. The combination unit is configured to adjust the high-resolution residue image according to a weighting factor and combine an adjusted high-resolution residue image and the high-resolution image, in order to output an enhanced image, wherein the weighting factor is different from a reference weighting factor being used in a deep-learning procedure for training the deep-learning residue network unit.

Growing community of inventors

Hsinchu, Taiwan

Yi-Ting Bao