Precise image semantic segmentation and optimization method using interaction means

Abstract

A precise image semantic segmentation and optimization method using an interaction means comprises the steps that interaction is achieved through marking and delimiting points input by a user, then semantic features are extracted through a deep convolutional network to achieve preliminary segmentation of a to-be-segmented image, and then a conditional random field model of the to-be-segmented image is constructed, the preliminary segmentation result is optimized by using the underlying features of the to-be-segmented image, and finally the accurate segmentation result is further optimized through an interaction means. The delimiting point is used as interactive input, the operation is simple and convenient, and the segmentation effect is improved; image semantics are extracted by using a full convolutional neural network, and a segmentation result has semantic integrity; the user is allowed to guide modification through manual intervention, and then image bottom layer features are used to assist in result optimization.

Description

technical field [0001] The present invention relates to a technology in the field of image processing, specifically a convenient and precise image segmentation and optimization method that uses demarcation points as an interactive means and integrates image underlying features and semantic information. Background technique [0002] Image segmentation is the process of dividing an image into several specific regions with unique properties and extracting objects of interest. It is a key technology in digital image processing. The existing interactive image segmentation technology requires users to provide part information through system interaction. The image information is extracted and processed by the algorithm, and the image is divided into two parts: the foreground that the user needs and the background that is not needed. At present, the more common information input methods include frame selection, smearing and clicking. Among them, frame selection can provide the most...

AI Technical Summary

Problems solved by technology

Among them, frame selection can provide the most abundant information, but at the same time, it cannot remove redundancy. Users use a rectangular frame to specify the area where the target object exists, and depending on the shape and type of the object, a large number of backgrounds may appear in the frame at the same time. In order to reduce the interference of irrelevant information, users can use the method of smearing to indicate a specific area as a reliable foreground, which ensures the accuracy of the information while providing a large number of foreground areas, but the size of the smeared area will vary. To a large extent, it affects the segmentation effect and requires cumbersome interaction. It is difficult to achieve the desired effect when there are many object blocks or rich textures; the click method provides less accurate information and is a relatively convenient interaction method. Used for the specification of segmentation boundaries or further optimization of segmentation results

The three interaction methods have their own advantages and disadvantages. A single type of input often cannot meet the actual needs. The more common image segmentation methods use a variety of interactive methods. For example, GrabCut uses frame selection to roughly determine the area where the object is located. And allows users to specify the foreground and background by smearing to optimize the segmentation results; LazySnapping uses smearing to specify front and back background pixels, and uses click input to edit the boundary

[0003] The technical problems that need to be solved urgently in the existing image segmentation methods are: 1) the lack of high-level semantic information in the traditional image segmentation methods

3) Existing means of interaction are cumbersome and inefficient

Images