An Image Processing Algorithm to Find New and Old Bruises

Abstract. Stored ‘Red Delicious’ apples can have old and new bruises. Previous studies have shown the near-infrared (NIR) reflection of bruised tissue with respect to unbruised tissue to be darker in new (24 h) bruises and lighter in old (four to six weeks) bruises. An algorithm was developed to identify both old and new bruises in the same NIR image. Apple images were low pass filtered to remove small surface characteristics (lenticels) and normalized. Absolute difference between these images and a standard (delrin sphere) representing unbruised tissue were also low pass filtered. A binary image of bruised/unbruised tissue was formed based upon a user defined threshold level. Shapes of the bruised areas were evaluated and, based upon a shape factor, either dismissed or counted as bruised areas. The algorithm detected both old and new bruises except for large 25 to 35 mm (0.98 to 1.38 in.) diameter old bruises. These bruises had areas within their boundaries where NIR reflectance had not changed completely from darker to lighter than unbruised tissue. This patterning altered the bruise shape causing the algorithm to discount them.