The Lab color model was selected as the segmentation color space. The developed system was able to classify each pixel by calculating the smallest Euclidean distance between the pixel and a set of color markers. The RGB images were taken at an early stage of the experiment, and then converted to Lab images. Small regions, which include some regions from the background and the hens, were chosen. The average color of each region was calculated to segment the hen and the background. For automatically separating overlap hens and counting the number of hens in group-housed environments, an algorithm based on ultimate erosion and concavity seek was used to provide the most accurate results. With the results of 108 images, it showed that the algorithm was able to achieve an accuracy of 93.5% for counting the number of hens in image, and an accuracy of 89.8% under conglutination condition. 