Object counting is an important problem in image analysis. All the proposed techniques until now need to split someway the conglomerates of objects in the image before determining their number. The improved technique here proposed uses the skeleton of the image to accomplish the same task. The original technique was first introduced in [1]. The improved technique also uses the number of terminal points, NTps (points with just one neighbor) and the number of three-edge-points, NTEps (points with just three neighbors) in the skeleton to compute the desired number of objects in the image. The approach is applicable to the case of objects such that when its skeleton is obtained the sum of (NTps+NTEps)=2. The technique can efficiently handle noisy branches in the skeleton and the undesired decomposition of crossing points into three-edge-points normally present when an image is skeletonized. This was not done for the original technique. Both the original and the improved technique are compared in several scenarios.