Efficient inscribing of noisy rectangular objects in scanned images

Objects identification in images is generally hard unless the objects are simple geometric shapes such as circles, rectangles, have very particular properties. Even simple geometric shapes can be hard to identify if they deviate even a little from the ideal. We examine the question of identifying and segmenting noisy rectangles, where edges may be ragged, corners may be missing and so on. We test the robustness of our scheme on receipts and small documents obtained from real scans.