Building a General Classification System for Image Quality Defects

Over the course of several years, many image-based intelligent solutions have been developed for a variety of use cases. The one thing they all have had in common is an array of image quality issues present in the raw image datasets used to build and test the solutions. According to Forbes (2016), "Data Scientists spend 80% of their time finding, cleaning, and trying to organize the data". This trend is further observed during cleaning image datasets in which human error is also prevalent. "A bad dataset will lead to a bad model" -- If the image quality defects are due to an error while capturing or does not represent natural life like conditions, then the model trained is sure to fail.