Maltodextrin is a substance that is being increasingly used to preserve the physicochemical and biological properties of many active compounds. Therefore, determining the particle size distribution (PSD) of the particles having maltodextrin matrices is a crucial issue to control their end-use properties. This can be done directly by laser diffraction (LD) in a dry way or via SEM image analysis. In this paper, a new method of segmentation of quasi-circular particles from grayscale images called curvature analysis method (CAM) is proposed. This method is compared to two other widely used methods: Circular Hough Transform (CHT) and Stochastic Watershed (SW). It aims in particular to reduce the drawbacks of these two methods: a large number of false detections and an inaccuracy with respect to the mean particle size in case of a large number of overlapping particles. The method is validated using synthetic images generated with a model allowing to simulate gray level images similar to the maltodextrin particle images from the SEM. The CAM method is then applied to real images and the resulting PSD is compared to the one provided by the LD technique. Overall, the results obtained by the CAM method are much better than those proposed by CHT and SW on synthetic images and than the LD method on real images.