Image processing method and device, electronic equipment and storage medium
A technology of image processing and electronic equipment, applied in the information field, can solve problems such as strange images, large image distortion, and unsatisfactory image quality, and achieve the effect of improving image quality
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
Method used
Image
Examples
example 1
[0187] The technical solution for this example consists of three parts:
[0188] 1. A convolutional neural network model for edge detection. For the input face picture, the model is responsible for obtaining an accurate face edge line detection result (such as outer eyelids, outer face contour lines, etc.).
[0189] 2. A conditional encoding and decoding network, from the contour information obtained in the first step, the structural representation is extracted through the network, as clear structural feature information to help the encoder decompose the appearance texture features and structural features of the input face image.
[0190] 3. A weight normalization and decoder design based on perceptual quality, which further helps to improve the generation quality.
[0191] The neural network provided in this example, by decomposing the spatial structure information of the picture, clearly decomposing the texture features and structural features of the encoder appearance, main...
example 2
[0193] This example provides an image processing method that includes:
[0194] Given an image x; then need to get x and The mapping relationship between G. The G may include: φ app and u str . The mapping relationship can be passed through the texture feature z=φ app (x,c) and y=u str (c).
[0195] The aforementioned deep learning model can be constructed using a conditional variance autoencoder (CVAE) network.
[0196] The aforementioned probability distributions and / or probabilities are solved using the following functional relationship.
[0197] logp(x / y)≥E q [logp(x / z,y)]-D KL [q(z / x,y),p(z / y)]; Based on this functional relationship, by solving the maximum value of p(x / y), q(z / x,y) and p(z / y) can be obtained ). Among them, q(z / x,y) can be approximated as p(z / y) 2 .
[0198] The network provided in this example can be trained with a function of the following stochastic objective:
[0199]
[0200] q φ (z / x,y) satisfies the distribution constraint N(0,I)....
PUM
Login to View More Abstract
Description
Claims
Application Information
Login to View More