<p>The accuracy of object recognition in binary images largely depends on the informativeness of the features used. In this study, object description is based not on a single contour-length value but on the distribution of contour lengths obtained from a set of images belonging to the same class. This representation preserves </p><p> the natural variability of object shapes within a class and exploits it during classification. The statistical model is constructed sequentially. First, the image is binarized and subjected to morphological processing. The largest external contour is then extracted, and its contour length is determined. The obtained contour-length </p><p> values form a sample used to estimate the empirical distribution and construct a histogram. Subsequently, the centers of the non-empty histogram bins are used as reference features, while their relative frequencies serve as weighting coefficients. Classification of a test image is performed using the nearest-neighbor method. </p><p> The proposed approach was evaluated on two experimental datasets containing 3,600 binary images of ten classes and 10,800 binary images, respectively. In all experiments, the statistical representation of contour lengths preserved class separability under object rotation. At the same time, the computational procedure </p><p> remained simple, and the classification results were stable. These characteristics indicate that the proposed method is promising for applications in computer vision, industrial inspection, technical diagnostics, and automated image analysis.</p>