Drawing object identification and extraction method in complex scene

Abstract

The present invention relates to the field of image identification and segmentation technology, especially to a drawing object identification and extraction method in a complex scene. The method comprises the following steps of: the step 1: constructing a detection and identification model, and employing an image set with object frames and object type marks to train the model; the step 2: collecting a scene image comprising user drawing content; the step 3: employing the trained detection and identification model to locate a drawing area in the image, and identifying a drawing object; and thestep 4: selecting the drawing area, and employing an image segmentation technology to extract a drawing object contour area. The drawing object identification and extraction method in a complex scenecan be applied in the field of children drawing teaching, can increase intelligent interactivity of the drawing process and can improve the drawing experience.

Description

technical field [0001] The invention relates to the technical field of image recognition and segmentation, in particular to a drawing object recognition and extraction method in complex scenes. Background technique [0002] Image recognition and segmentation is an important and fundamental problem in the field of computer vision, and it is also a challenging task. In recent years, the development of deep learning technology has made great achievements in the field of computer vision, especially in the field of image recognition, which has achieved results beyond human beings, and is also making efforts in end-to-end image segmentation. [0003] Painting recognition is very useful in the fields of children's painting education and children's cognitive enlightenment. However, unlike the recognition of natural images, paintings will be more abstract, especially children's paintings, which will be more concise and imaginative, which means that the images have less feature infor...

AI Technical Summary

Problems solved by technology

Jun Guo et al. proposed to use Gabor features to construct low-dimensional representations, and further use sparse coding to construct high-level features, and then use support vector machine classification methods to complete classification recognition [1]; Zhao Peng et al. proposed to use deep learning methods to complete painting recognition. In order to solve the problem of less information in stick figure painting, a method of increasing the size of the convolution kernel is proposed [2]; these methods are applied to public data sets that only contain painting images, although they achieve an effect close to that of a human being, but in practice In the application, complex scene information (non-painting content) cannot be processed; the Chinese invention patent "a painting tutoring method and device" [3] discloses a painting tutoring method, including the following steps: receiving user's painting information; According to the image recognition technology, the line outline in the painting information is extracted, and different foreground and background objects are identified; the thickness, smoothness, and length change rules are analyzed; the straight line detection is performed on the object, and the eye level and center point are identified; image processing and light analysis are carried out. Identify the highlights, shadows and shadows in the image; specify a painting style as a limiting condition, provide guidance to the user's painting information image, and give correction opinions; it can also only deal with the area on the drawing paper. However, when the camera captures the painting content , due to lighting, angles, shadows, distances, etc., when the acquired images are more complex, the recognition process cannot be completed accurately in scenes such as indoor living rooms, teaching classrooms, etc., so some technical methods are needed to complete the painting from these complex scenes Image recognition and segmentation

Currently, there is no way to accomplish this task

Images