[论文解读] K-means Segmentation Based-on Lab Color Space for Embryo Egg Detection.

Shoffan Saifullah|arXiv (Cornell University)|Mar 3, 2021

Food and Agricultural Sciences参考文献 33被引用 9

一句话总结

本研究提出在Lab色彩空间图像上使用K均值聚类分割技术，以检测蛋黄中的胚胎，实现早期孵化成功率预测。通过k=3将图像聚类为三个区域——背景、蛋和蛋黄，该方法基于颜色分割识别胚胎，随后进行灰度化增强和形态学处理，实现了高精度，MSE=0.0486，MSSIM=0.9979。

ABSTRACT

The hatching process also influences the success of hatching eggs beside the initial egg factor. So that the results have a large percentage of hatching, it is necessary to check the development of the embryo at the beginning of the hatching. This process aims to sort eggs that have embryos to remain hatched until the end. Maximum checking is done the first week in the hatching period. This study aims to detect the presence of embryos in eggs. Detection of the existence of embryos is processed using segmentation. Egg images are segmented using the K-means algorithm based on Lab color images. The results of the images acquisition are converted into Lab color space images. The results of Lab color space images are processed using K-means for each color. The K-means process uses cluster k=3, where this cluster divided the image into three parts, namely background, eggs, and yolk eggs. Yolk eggs are part of eggs that have embryonic characteristics. This study applies the concept of color in the initial segmentation and grayscale in the final stages. The results of the initial phase show that the image segmentation results using k-means clustering based on Lab color space provide a grouping of three parts. At the grayscale image processing stage, the results of color image segmentation are processed with grayscaling, image enhancement, and morphology. Thus, it seems clear that the yolk segmented shows the presence of egg embryos. Based on this process and results, K-means segmentation based on Lab color space can be used for the initial stages of the embryo detection process. The evaluation uses MSE and MSSIM, with values of 0.0486 and 0.9979; this can be used as a reference that the results obtained can indicate the detection of embryos in egg yolk.

研究动机与目标

• 改进鸡蛋中早期胚胎检测以提高孵化成功率。
• 解决在孵化第一周内识别胚胎发育的挑战。
• 开发一种基于视觉胚胎特征的可靠、自动化方法，用于区分可育鸡蛋。
• 评估基于Lab色彩空间的颜色分割在胚胎检测中的有效性。

提出的方法

• 获取鸡蛋图像并将其转换为Lab色彩空间，以增强颜色对比度，便于分割。
• 使用k=3的K均值聚类将图像分割为三个区域：背景、蛋和蛋黄。
• 基于颜色和纹理特征，将蛋黄区域用作胚胎存在的指示。
• 对分割后的图像进行灰度化转换、图像增强和形态学操作，以优化胚胎检测效果。
• 利用MSE和MSSIM指标评估分割精度和图像保真度。

实验结果

研究问题

• RQ1 在Lab色彩空间上使用K均值聚类分割能否有效区分含胚胎的蛋黄与其他蛋成分？
• RQ2 与标准RGB色彩空间相比，Lab色彩空间转换如何提升分割精度？
• RQ3 后分割处理在多大程度上增强了蛋黄中胚胎特征的可见性？
• RQ4 MSE和MSSIM等定量指标如何体现所提出胚胎检测方法的可靠性？

主要发现

• 使用k=3的K均值聚类成功将鸡蛋图像分割为三个明显区域：背景、蛋和蛋黄。
• 经过灰度化和形态学处理后，分割出的蛋黄区域清晰显示出胚胎的存在。
• 该方法实现了极低的均方误差（MSE）0.0486，表明图像重建具有高保真度。
• 结构相似性指数（MSSIM）高达0.9979，证实分割输出与原始图像结构高度一致。
• Lab色彩空间与K均值聚类分割的结合，可实现对蛋黄中胚胎的精确、早期检测。