Sketch-based interactive iteration type virtual shoeprint image generation method

Abstract

The invention provides a sketch-based interactive iteration type virtual shoeprint image generation method. The method comprises the following steps: constructing an overall network architecture; wherein the overall network framework comprises a hole residual multi-level fusion full-convolution shoeprint image generation network, a full-convolution deep and shallow feature fusion shoeprint image discrimination network and a VGG19 pre-trained on an Image Net; performing offline model training on the constructed overall network; and generating an online virtual shoeprint image based on the offline trained model. According to the method, the shoeprint sketch is used to realize the generation of the shoeprint image, and the problems that the background is complex and the field shoeprint image is difficult to extract, which cannot be solved by the existing traditional image restoration algorithm and deep learning, are solved. The criminal investigation personnel are assisted to utilize crime scene information as much as possible to carry out case detection, and the case solving efficiency is improved.

Description

technical field [0001] The invention relates to the technical field, in particular to a sketch-based interactive iterative virtual shoe print image generation method. Background technique [0002] At present, sketch-based image generation is mainly divided into two categories. The first category is based on traditional image correlation algorithms. The second category is the image generation algorithm based on deep learning. The specific content of each method is as follows: (1) The traditional image sketch generation algorithm adopts the method of search fusion, first searches the image blocks related to the sketch from the large-scale image database, and then fuses the image blocks. (2) In recent years, with the development of deep learning, generative adversarial networks have been increasingly applied to image generation. Isola et al. proposed a conditional generative network model with supervised training, but this model is only suitable for dense images, and the eff...

AI Technical Summary

Problems solved by technology

Isola et al. proposed a conditional generation network model with supervised training, but this model is only suitable for dense images, and the generation effect of sparse images such as input sketches is not satisfactory.

[0003] Whether it is the traditional sketch-based image generation algorithm or the deep learning sketch-based image generation algorithm, it has good generation ability for complex and complete sketches, but it cannot produce good generation results for too simple and sparse sketches. , the main reason is that there are too few areas that can be used in a picture, and the content information and texture information are seriously insufficient, which cannot satisfy the inference ability and learning ability of the algorithm, so the generated image will be too smooth to satisfy the subjective feeling of the human eye. , inaccurate matching, inconsistent textures, lack of detail information, etc.

Especially for shoe print images, the sole pattern has rich line and texture information, and it is particularly cumbersome to draw, so it is urgent to give an algorithm that can generate a complete shoe print image with the simplest sketch

[0004] In the on-site shoe print images, the background is often too complex and the shoe print images are difficult to extract.

This greatly reduces the usefulness of the evidence

Images