Enhance reliability of script in identifying an image exists in a digitised document (OpenCV)

I'm trying to use existing software on a project to extract printer's illustrations/ornaments/ornate letters from digitised books. To do this, I have turned to [login to view URL] which draws on OpenCV. However, the software does not reliably identify all of the images on a page; in some cases it does not recognise an illustration exists at all. Can the settings in the [login to view URL] file (attached) be changed to significantly improve the chances of identifying/extracting illustrations? For instance, [login to view URL] - the illustration is identified as something to be extracted. But [login to view URL] the image is not seen. Also, in [login to view URL] only one image is found (the ornate letter on the left) but not on the right. [login to view URL] is not seen as containing an image. [login to view URL] - software does not see any illustrated elements (though there are two). [login to view URL] no illustration is detected.