The perimeter of a target boundary in a 2D image is an essential object feature in image analysis. However

this feature is usually inaccurately estimated because of discontinuous or blurred target boundaries. Accordingly

this study proposes an improved method of perimeter estimation. This method is based on gray-level information and combined with boundary tracking. Different from the traditional methods that generally use binary information to calculate the perimeter

the proposed method utilizes gray-level information in digital images to obtain substantial information on a target boundary. The concepts of pixel coverage

internal boundary

and external boundary are introduced. The slope at each configuration on the internal and external boundaries is computed

and the perimeters of the internal and external boundaries are estimated based on the arc length integration formula. The perimeter of the target object boundary is obtained by combining the perimeter information on the internal and external boundaries. The perimeters of the target object boundaries are estimated using the proposed method and three classical methods. Synthetic images are used to obtain the exact perimeters of the target objects. The results are compared with the ground truth perimeters of the target objects in synthetic images. In the first two experiments

synthetic images with continuous and blurred boundaries are tested. The performance of the proposed method is similar to that of the classical methods but is better than that of the original gray-level method. In the third experiment

synthetic images with discontinuous and blurred boundaries are tested. In this case

the two classical geometrical methods are not applicable

whereas the proposed method remains to have satisfactory performance. The advantage of the proposed method is more obvious when the boundary of the target is complex. Compared with the classical methods

the proposed method has better adaptability and stability for target objects with blurred or discontinuous boundaries.