Objective: Color image segmentation is an important image analysis technology, which has important applications in image recognition system. The quality of image segmentation directly affects the effect of image processing. However, because of the noise, the uneven color and the weak boundary, color images in real life are usually difficult to be segmented precisely. In this paper, we proposed a watershed segmentation algorithm based on the homomorphic filtering and morphological hierarchical reconstructions. By combining the advantages of homomorphic filtering, morphological operations and watershed transform, the qualities of the color images’ segmentations are improved. Method: The watershed transform algorithm has been widely used for segmentations of images, because of its advantages such as low computational burden, high accuracy and continuous extraction. However, due to the fact that there are a lot of irregular regions and noises in the image, segmenting the image only relying on a watershed transform algorithm is easy to form a large number of false contours. In order to improve the quality of image segmentation by watershed transform, we get help from the homomorphic filtering and the morphological reconstruction. Firstly, the proposed algorithm used the “sobel” edge operator to compute the gradient of each color component according to the image’s R, G and B values, and the maximum value was selected as the gradient of the color image. After the gradient map of a color image was extract, it was modifies by the homomorphic filtering using Fourier transform. The filtering helps to highlight the foreground contour information on one hand and removes the detail texture noise on the other side. Since there were still some irregular details and noise in the gradient image after filtering, especially at the boundary and background, and the morphological reconstruction operators were able to improve this shortcoming, the modified gradient map was then reconstructed hierarchically by using the operators of open and close morphological reconstructions. According to the cumulative distribution function of the gradient map and the distribution information of the gradient histogram after filtering, the formula for calculating the number of gradient layers was given and the sizes of morphological structure elements, which were decreasing with the increase of the gradient value in each layer, were then calculated adaptively. Finally, the algorithm applies the standard watershed transform to the reconstructed gradient map, and the image segmentation was realized. Result: In order to verify the effectiveness of the algorithm, this paper select four color images of different features to segment in the experiment. Results indicate that the proposed algorithm can effectively restrain the over segmentation and keep the weak boundary, hence the segmentations are more accurate compared with other watershed algorithms. Furthermore, for objectively evaluating the performance of different segmentation methods, this paper quantified the experimental results by unsupervised evaluation of segmentations, which applied the synthesize index combined with regional consistency and diversity indexes. The evaluation index values of our algorithm in the four test images are 0.6333、0.6656、0.6293 and 0.6484 respectively，higher than the results other watershed algorithms, meanwhile the segmentation performances are also better. From the point of view of timeliness, this algorithm takes a little longer time, but it has little difference with the other two algorithms. Conclusion: The watershed transform is a widely used algorithm for image segmentation, but it often leads to over segmentation. Many methods have focused on solving this problem, while ignoring the weak boundaries of images, which are also important in segmentations for application. This paper proposed a new improved watershed algorithm for color images. In the algorithm, the homomorphic filtering is used to preserve the weak boundary of the image, and an adaptive morphological reconstruction is applied to suppress the over segmentation of watershed transform. A balance is found between under segmentations and over segmentations. Segmentation results of the algorithm are closer to the human perception of the images. No matter from the evaluation index or segmentation performance, the proposed algorithm performs better. This algorithm is not sensitive to noise and has good robustness with widely application in computer vision, traffic control, biomedical and other aspects of the target segmentations.