Highlight Feature Detection And Its Application In Fruit Classification

Highlight is a region of color distortion on the observed surface of material nearest to light source.It’s generally available in images as long as visible light exists,and often be eliminated as noise in image processing.However,highlight,distinct from overexposure,only suppresses texture caused by color on the surface.As it does so,it emphasizes the brightness vibration caused by the uneven characteristic of each material.So,the highlight feature could play an extinct role in material classification.This topic fully utilized the advantages of highlight feature to proposed the concept and algorithm of highlight feature detection,and applied it to the area of material classification,especially fruit classification.The specific research content is as follows:1、Platform construction of image acquisition.Using SLR to photograph materials in different color and intensity of illumination.Balance between light conditions and SLR parameters would make it that the highlight feature is always clear without overexposure.2、Propose HIGHLIGHT illumination model.Classical illumination models were analyzed on their principles and applicable scenes.The proposed HIGHLIGHT illumination model emphasized local brightness change,which couldn’t be represented by the classical one.By using the model in experiment,we proved its applicability on highlight detection.3、Highlight feature detection and its application on fruit classification.Highlight feature was extracted by HIGHLIGHT illumination model,and then described as a vector by STFFT algorithm.For its application on fruit classification,we improved DBSCAN cluster algorithm with max-win-voting strategy and input highlight descriptor into it.The experimental result is excellent with up to 100% accuracy.