Invariant Image Recognition by Zernike Moments
IEEE Transactions on Pattern Analysis and Machine Intelligence
| Hi-index | 0.00 |
We describe a novel system for grading oranges intothree quality bands, according to their surfacecharacteristics. The system is designed to processfruit with a wide range of size (55–100 mm), shape(spherical to highly eccentric), surface colorationand defect markings. This application requires bothhigh throughput (5–10 oranges per second) and complexpattern recognition. The grading is achieved bysimultaneously imaging each item of fruit from sixorthogonal directions as it is propelled through aninspection chamber. In order to achieve the requiredthroughput, the system contains state-of-the-artprocessing hardware, a novel mechanical design, and three separate algorithmic components. One of the keyimprovements in this system is a method forrecognising the point of stem attachment (the calyx)so that it can be distinguished from defects. Aneural network classifier on rotation invarianttransformations (Zernike moments) is used to recognisethe radial colour variation that is shown to be areliable signature of the stem region. The successionof oranges processed by the machine constitute apipeline, so time saved in the processing of defectfree oranges is used to provide additional time forother oranges. Initial results are presented from aperformance analysis of this system.