Given the variation of images across the collection, it is very likely that we will need to tune various elements of the pipeline for images with certain characteristics. If we could identify these characteristics (contrast, script type, layout density, etc.) on the fly, we could set parameters automatically rather than needing to hard-code them for specific batches of images. Based on the results of testing with the representative image set, we should determine which image characteristics are most important in determining pipeline performance and develop function(s) to quantify these characteristics so that optimal parameters for a given image can be determined on the fly.
Given the variation of images across the collection, it is very likely that we will need to tune various elements of the pipeline for images with certain characteristics. If we could identify these characteristics (contrast, script type, layout density, etc.) on the fly, we could set parameters automatically rather than needing to hard-code them for specific batches of images. Based on the results of testing with the representative image set, we should determine which image characteristics are most important in determining pipeline performance and develop function(s) to quantify these characteristics so that optimal parameters for a given image can be determined on the fly.