Experimental study on the impact of endoscope distortion correction on computer-assisted celiac disease diagnosis

The impact of applying barrel distortion correction to endoscopic imagery in the context of automated celiac disease diagnosis is experimentally investigated. For a large set of feature extraction techniques, it is found that contrasting to intuition, no improvement but even significant result degradation of classification accuracy can be observed. For techniques relying on geometrical properties of the image material (“shape”), moderate improvements of classification accuracy can be achieved. Reasons for this somewhat unexpected results are discussed and ways how to exploit potential distortion correction benefits are sketched.