Local defects are very common on printed pages. Automatic detection of such defects will help the product support personnel to diagnose the problem and fix it more efficiently. Among previous works on local defect detection on printed pages, most of them divide the printed page into small blocks and calculate the variation within each block. This method is time consuming and not robust in dealing with defects at different scales. In this paper, we propose a robust framework for detecting the local defects on scanned printed pages. To achieve the efficiency and robustness, our framework applies the Gaussian pyramids method and the selective search method. We also create manual features for classification to increase the detection accuracy. Finally, applying our method on printed pages demonstrates its efficacy.