GrabCut improvement-based image segmentation method
An image segmentation and image technology, which is applied in image analysis, image enhancement, image data processing, etc., can solve problems such as ignoring color features, low segmentation accuracy, and complicated Haar wavelet parameter setting, so as to reduce error rate and improve operation Efficiency, the effect of increasing the Kappa coefficient
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
Method used
Image
Examples
Embodiment 1
[0072] Such as figure 1 As shown, an improved image summing algorithm based on GrabCut includes the following steps:
[0073] Execute step 100, pre-segmentation of multi-scale watershed.
[0074] The watershed algorithm is a classic segmentation algorithm, which can not only preserve the edge of the original image well, but also ensure that the difference of each small area is small enough. However, due to quantization errors, object details, noise, etc., it is easy to cause over-segmentation.
[0075] In view of the shortcomings of over-segmentation and poor edge details, this paper uses the watershed algorithm based on multi-scale morphological gradient operators to preprocess the image. The traditional morphological gradient operator is shown in the following formula:
[0076]
[0077] in: and ⊙ represent expansion and erosion operations, respectively, and B is a structural element. The above formula is also called single-scale morphological gradient operator, and ...
Embodiment 2
[0104] Such as figure 2 Shown, the step of the improved GrabCut algorithm among the embodiment 1 is as follows:
[0105] Step 200 is executed to preprocess the image. Input the image I, perform the second watershed pre-segmentation for I, and use the color mean value of the obtained small area as the node of the pixel for subsequent processing.
[0106] Step 210 is executed to initialize the image.
[0107] Execute step 220 to perform iterative minimization.
[0108] Execute step 230, target output. Obtain a new a=0, a=1 pixel set, and output the image pixels with a=1 to achieve the foreground target output.
Embodiment 3
[0110] Such as image 3 As shown, the steps of initializing the image in Embodiment 2 are as follows:
[0111] Execute step 300, non-complete numbering, the user sets the background T B Transform the ternary graph T into a binary labeling problem. User initial interaction only needs to determine T B , leaving the foreground blank, ie T U Take the complement of the background, that is,
[0112] Execute step 310, for all background pixels, set their transparency a to 0, that is, a=0; for the unknown area T U , let a=1.
[0113] Execute step 320, for the two collections of a=0 and a=1, use the k-means clustering method to initialize the GMM of foreground and background, and obtain the GMM parameters (π k , u k , Σ k ) initial value.
PUM
Login to View More Abstract
Description
Claims
Application Information
Login to View More