The 2010 ASHS Annual Conference

Estimates of the extent of coverage of red blush on apple surface play a role in the evaluation of apple strains, varieties, and horticultural practices. These estimates, however, may be subject to differences in ability or skill of observers. As such, they are data of unknown quality. This study was designed to quantify the uncertainty in visual apple blush estimates of 12 strains of Honeycrisp by trained observers, by comparing their estimates to output from digital image analysis (DIA) techniques. Experts were surveyed with images of apples displaying partial blush and asked to choose black/white representations of the blush. Preferred representations of the blush/non-blush threshold for Honeycrisp apples approximated a hue value 56 in CIELUV L*C*h color space. Using this threshold and two others representing the approximate bounds of expert opinion, approximately 3000 images of Honeycrisp apples harvested in 2009, taken with two digital cameras under controlled lighting and camera settings, were analyzed. The images were processed with color management techniques and cross-referenced with standard color chips photographed in the light box and with output from a CM2600d Minolta spectrophotometer used to measure the color chips and selected apples. Pixels in each image were categorized as blush, non-blush, and background. The background was ignored and percentage blush was calculated for each image. Frequency distribution curves of percentage blush were prepared for each of the blush threshold values examined -- the preferred value, the lower bound, and the upper bound. Percentage blush as a function of frequency distribution approximated the exponential recovery function, indicating a strong tendency toward higher blush for the 2009 season. Experts were asked to rate the percentage blush of 99 images of these apples, representing 9 randomized replications of each of 11 blush percentage classes. Experts estimated blush in close correspondence to DIA (R²=0.96), but showed slight systematic bias. DIA proved to be a viable methodology for examining large volumes of photographic data accurately, with higher consistency than human judgment, and with minimal effort. The authors wrote two macros in the open-source software package ImageJ and have made them available to the research public. The first macro is an apple isolator that modifies a digital image of an apple by removing the background (low saturation) portion of the image. The second macro calculates percentage blush of an image of an apple using a threshold hue value specified by the user.