Fast apple inspection in colour space

Abstract

This work presents an apple image library of over 20,000 colour images of four varieties of randomly oriented apples (both mono–coloured and variegated varieties). The image library is offered to other researchers in apple inspection. The images were analysed using a colour difference operator, a grey level difference operator and an islanding routine to identify points of interest that could be calyxes, stems or blemishes. The method correctly identified 98% of calyxes, 93% of stems and 83% of defects. 5% of blemish–free images were mistakenly identified as having defects; the 'ghost islands' arising because the natural variation in the peel colour of variegated apples was sufficiently pronounced to look like a defect. Image processing time is fast enough for implementation in commercial packing lines.