An Automated Digital Image Analysis Technique for Quantifying Apple Blush Coverage

Estimation of red blush coverage extent is important in evaluation of apple appearance. USDA grading standards, for example, specify percentages of blush coverage necessary to meet three grades for several varieties. Also, consumer choice is influenced by blush as it informs visual estimations of expected fruit quality. Researchers and industry professionals have conventionally relied on visual estimates of blush by trained observers. These estimates, however, may be subject to differences in skill and may not always be consistently applied. We present an alternative method for quantifying blush, applying image analysis techniques to digital photographs. Approximately 3,000 images of Honeycrisp apples from the 2009 growing season were photographed under a variety of conditions and analyzed. Issues in color accuracy, image consistency, automation of analysis using macros in Java (ImageJ), lighting, and camera operation were explored. The resulting method consists of a process to create accurate blush estimates for digital images of apples. The first step is instrument calibration. This includes building a light booth appropriate for apple images, establishing camera settings appropriate for light conditions, and analyzing the error of camera output by comparing red-green-blue values from images of captured reference color chips to published values and to the output of spectrophotometer measurements. Once the error is understood, appropriate linear transforms for RGB vectors can be applied to output images for data correction. Once the accuracy and reliability of the system has been maximized, images of the apples are taken in the light booth.  Images are pre-processed to remove the image background by re-assigning the values of pixels with saturation below a given threshold to null. Another threshold is then applied to the resulting image in order to establish the hue boundary between blush and non-blush pixels. Images are then reclassified to values representing colors below the blush threshold, above the blush threshold, and null values for non-apple pixels. Analysis consists of analyzing pixels classified as blush and non-blush, and calculating the prevalence of each class to the whole. The process is then automated so that the pre-processing and classification steps are expressed in two Java macros that can be automatically applied to all images in a given directory in a matter of minutes. The Java macros operate within the open-source free software program ImageJ, available from the National Institutes of Health. The authors will also make the apple-blush-analysis macros freely available to the public.