We suggest a method to detect mine-like objects in side scan sonar images. First, we use the well known K-means algorithm recognizes which recognizes the presence of mine-like segments, followed by the Chan-Vese active contour algorithm to sharp and restore the edges of the suspected segments. Then, we convert the image into a shadows and highlights map based on geometrical features of the connectivity between shadows and highlights. By exploiting the unique geometrical features of mine-like objects, we implement a Neyman-Pearson test to detect the mine-like objects given the connectivity map while reducing false alarms and other artifacts. Testing on real data images show that the suggested approach have good detection results.