研究目的
To develop an image thresholding approach that effectively handles noisy and uneven grayscale images by using a Gaussian mixture model and incorporating neighborhood information.
研究成果
The proposed image thresholding method based on GMM and EM algorithm, with neighborhood information in posterior probabilities, is effective for noisy and uneven grayscale images, outperforming several traditional methods in accuracy and robustness. Future work should focus on improving boundary accuracy and extending to multilevel thresholding.
研究不足
The method may achieve poor results for images with thin small targets, and boundaries can be blurred. It is primarily designed for bi-level thresholding and may not handle multilevel segmentation efficiently without extensions.
1：Experimental Design and Method Selection:
The method uses a Gaussian mixture model (GMM) with two components for foreground and background, employing the expectation maximization (EM) algorithm for parameter estimation and Bayesian criteria for binary map generation. Neighborhood information is considered in posterior probability calculations to improve noise robustness.
2：Sample Selection and Data Sources:
Synthetic images with added Gaussian and Salt & Pepper noise, uneven grayscale synthetic images, and real images are used for testing.
3：List of Experimental Equipment and Materials:
A computer with Intel Core i5 2.53 GHz CPU, 8GB memory, Windows 7 operating system, and MATLAB R2015a software.
4：53 GHz CPU, 8GB memory, Windows 7 operating system, and MATLAB R2015a software.
Experimental Procedures and Operational Workflow:
4. Experimental Procedures and Operational Workflow: Steps include initializing parameters using mean gray value, iterating EM algorithm (E-step and M-step), calculating posterior probabilities with neighborhood effect, and generating binary images based on Bayesian criteria.
5：Data Analysis Methods:
Performance is evaluated using misclassification error (ME) and Dice metric (DSC), with comparisons to other methods like Otsu, Niblack, Otsu_2D, and Sauvola.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
