CN108765428A - A kind of target object extracting method based on click interaction - Google Patents

Abstract

The invention discloses a method for extracting a target object based on click interaction, which is referred to as click-cut for short. In order to simplify the interactive behavior in the interactive segmentation and make the interactive image segmentation simpler and automatic, the invention discloses a new interactive image segmentation method. In order to select the target area in the image, the user only needs to input a point in the desired area, and realizes the Click-cut method that can segment the foreground part of the image with a little user interaction. This method divides the image into the foreground area containing user interaction and the rest of the background area. This method can be roughly divided into two parts. First, the image is automatically divided into several super-regions, and then the region required by the user is determined according to the user interaction point using the region selection strategy. The invention has the advantages of simple operation, excellent performance, clear outline of the extracted target object, and can meet the demand for various new functions in the future, and can be applied to various occasions.

Description

A Target Object Extraction Method Based on Click Interaction

technical field

The present invention relates to a method for extracting an area of interest (target object) in an image through user interaction, and more specifically relates to an image segmentation and recognition method for extracting a foreground area in an image by a user inputting a click action, which belongs to image processing technology field.

Background technique

Image segmentation is the technology and process of dividing the image into regions with different characteristics and extracting the target of interest. It is a key step from image processing to image analysis. The research on image segmentation technology has a history of several decades, but so far people cannot find a general method that is suitable for all types of images. Automatic segmentation technology is the most common way. Automatic segmentation is to develop a method to automatically segment images into various contours without user intervention. In recent years, many automatic segmentation algorithms have been proposed. However, the segmentation results obtained by automatic segmentation technology are not accurate enough, and users cannot extract the target area according to their own wishes. Due to performance limitations, they cannot meet user needs.

In order to solve this problem, an interactive image segmentation method with user intervention has emerged. This method avoids the automatic segmentation of the machine. Interactive image segmentation integrates user interaction actions into image segmentation, in a supervised The segmentation method can effectively extract meaningful foreground regions in the image. Typical interaction methods include using a brush to draw a few strokes on the foreground and background, and drawing a box around the foreground. But these interactive methods all need to input different foreground and background brushes, and require users to estimate the distribution of foreground and background regions well.

Because the existing interactive segmentation methods require complex brush interaction information, such as an interactive image segmentation method in patent number "201310548279.2", or need to input a rectangular frame selection, such as an interactive image segmentation method in "201310587120.1", etc. Wait. All these technologies are very inconvenient to use. How to further simplify the user interaction information, such as the user only needs to input a little to get the target area to be extracted. There is no such method available yet.

The present invention proposes a new interactive segmentation method. In order to select the target area in the image, the user only needs to input a point in the desired area, and realizes a click interaction to segment the image. The present invention defines it as the Click-cut method .

Contents of the invention

In order to solve the problem of complex interactive information in the existing interactive segmentation methods, the main purpose of the present invention is to provide a very simple interactive image segmentation method, which only requires the user to input a point in the desired selection area, and the The image is segmented into a foreground region containing user interaction information and the remaining background region.

In order to achieve the above-mentioned purpose, the technical scheme of an interactive image segmentation method based on user interaction points in the present invention is as follows: assuming that an image to be processed is given, the image is first automatically segmented, and then the foreground is extracted according to user interaction. Implemented a little image interactive segmentation (Click-cut).

The technical implementation of the present invention includes 2 steps, first, the initial automatic division of the image: first, according to a pair of images, it can be preprocessed first, such as performing superpixel segmentation, how to combine the color features and texture features of the image to extract each superpixel respectively The color features and texture features of pixels, etc., construct a new superpixel similarity matrix, and then use an automatic community division method to divide the image into several community areas. After the automatic segmentation of the image is completed, this is completed The first step in the method of invention. Next, the inventive area selection strategy needs to be adopted: the user only needs to input a point in the area to be selected, and the area selection strategy will merge the areas near the user interaction point, output the area required by the user, and divide the image into foreground and background parts. Finally, the goal of obtaining the foreground or the target object is achieved with only a little interaction.

The beneficial effects of the present invention are:

The invention reduces the interaction behavior in the user interactive segmentation to the simplest behavior, that is, the user only needs to input a little, and the target area of the image can be extracted well, and the outline information of the target object can be obtained well. Based on the method of the invention, Click-cut image interactive processing software can be developed, or the image interactive processing function of Click-cut can be integrated in existing image processing software.

Description of drawings

Fig. 1 is a general frame diagram of the present invention.

Fig. 2 is a functional diagram of the automatic image segmentation part of the present invention.

Detailed ways

The present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments, but it is not intended to limit the present invention.

As shown in Fig. 1 and Fig. 2, concrete realization of the present invention: in Fig. 1, given original image (1) is as the input of interactive segmentation algorithm, and this image can be that direct camera equipment collects static picture or extracts from video file The frame-by-frame image is obtained, and then the image is automatically segmented, and the image format is not limited, it can be in jpg or bmp format (2). An image can be divided into a small number of regions according to the self-similarity of pixels. Generally, an image is finally divided into more than a dozen regions (3). The user can input a point on the target area to be extracted from the original image, or input a point on the image that has been automatically segmented into several areas (4). Then according to the area selection strategy, the area where the user’s point position is located, such as B1, is obtained, and all neighboring areas T of B1 are searched. If the area with the smallest similarity between a certain neighboring area Tj and all adjacent areas is B1, it means that Tj and B1 Most similarly, the Tj region is also added to B in the foreground region. This strategy is performed recursively until all regions in set B have been searched once. Finally, the region in B is classified as the foreground region, which is the target object to be extracted, and the remaining regions are classified as the background region (6).

In the embodiment shown in Fig. 2, the original image is pre-segmented, and some ready-made techniques can be used in this step, such as superpixel segmentation (1). An image contains a lot of pixels, and it takes a long processing time to directly use pixels as graph nodes to divide the network. Therefore, the image is over-segmented using an initial segmentation method, such as a superpixel segmentation method. After the initial segmentation, many small superpixel regions of a graph are obtained. Each region can be regarded as a node of the network. In order to use the graph cut method to automatically segment the superpixel region, it is necessary to build the connection relationship of each region , some features need to be used to describe each region, such as color features, edges, texture features, and geometric features, where color features and texture features can effectively represent the characteristics of the target (2). The present invention comprehensively utilizes these two features, and forms the relevant connection relationship of each superpixel region of the image through combination, that is, obtains a unified similarity matrix (3). Reprocessing contains thousands of superpixels, that is, dividing the network containing thousands of nodes, and dividing the image into only dozens or a few regions. Since we don't know the number of graphs to be divided, we need to use an automatic graph division method, such as the community division method (4). After a community division, an image is divided into regions with fewer superpixel regions. Since only one community division is performed, it may still contain at least dozens of superregions, and the network division process (5) is repeated. When the number of regions no longer changes or the ideal division region has been reached, the automatic segmentation method (6) is terminated.

The series of detailed descriptions listed above are only specific descriptions for feasible implementations of the present invention, and they are not intended to limit the protection scope of the present invention. Any equivalent implementation or implementation that does not depart from the technical spirit of the present invention All changes should be included within the protection scope of the present invention.

Claims (8)

1. a kind of based on the target object extracting method for clicking interaction, which is characterized in that be made of two parts, first automatically by image Division there emerged a several regions, and the number in region is much smaller than number of pixels, and interbehavior is clicked further according to one input by user, It determines the coordinate position clicked and is partitioned into the target object needed for user, i.e. foreground part.

2. according to claim 1 a kind of based on the target object extracting method for clicking interaction, which is characterized in that specific mistake Journey：It according to a given sub-picture, is first pre-processed, pre-segmentation and feature extraction, according to the feature construction of extraction at similar Matrix N * N are spent, the similarity matrix of N*N is divided into K nonoverlapping image-regions, wherein K automatically using Automated Partition Method Much smaller than N；User inputs a click interbehavior, by the location of pixels of user's interaction point, according to the region in this method Selection strategy merges near zone where user's interaction point, and the targeted object region extracted needed for output user will scheme As being divided into foreground and background part.

3. according to claim 1 a kind of based on the target object extracting method for clicking interaction, which is characterized in that described to carry It is Click-cut to take method, i.e. image a bit (Click), i.e., is directly divided into target prospect and background two by user's input Divide (cut).

4. according to claim 1 a kind of based on the target object extracting method for clicking interaction, which is characterized in that if figure It, can be simultaneously every when user thinks to extract these target objects simultaneously as in when target object disjunct there are several independences Input a little carries out Click-cut on a target object, while obtaining extracting these target objects.

5. according to claim 2 a kind of based on the target object extracting method for clicking interaction, which is characterized in that described Image segmentation is not limited to certain picture format, and any still image is all suitable for Click-cut.

6. according to claim 2 a kind of based on the target object extracting method for clicking interaction, which is characterized in that the spy Sign extraction includes color characteristic and textural characteristics.

7. it is according to claim 2 it is a kind of based on click interaction target object extracting method, which is characterized in that it is described from Dynamic division methods are automatic group dividing method.

8. a kind of based on the target object extracting method for clicking interaction according to claim 1-7 any one of them, feature exists In target object extracting method can be used in developing at Click-cut image interactions in the image clicked based on user The software of type is managed, or integrates the image interactive process function of Click-cut in conventional images processing software.