Image analysis of representative food structures: application of the bootstrap method

J Food Sci. 2009 Aug;74(6):R65-72. doi: 10.1111/j.1750-3841.2009.01220.x.

Abstract

Images (for example, photomicrographs) are routinely used as qualitative evidence of the microstructure of foods. In quantitative image analysis it is important to estimate the area (or volume) to be sampled, the field of view, and the resolution. The bootstrap method is proposed to estimate the size of the sampling area as a function of the coefficient of variation (CV(Bn)) and standard error (SE(Bn)) of the bootstrap taking sub-areas of different sizes. The bootstrap method was applied to simulated and real structures (apple tissue). For simulated structures, 10 computer-generated images were constructed containing 225 black circles (elements) and different coefficient of variation (CV(image)). For apple tissue, 8 images of apple tissue containing cellular cavities with different CV(image) were analyzed. Results confirmed that for simulated and real structures, increasing the size of the sampling area decreased the CV(Bn) and SE(Bn). Furthermore, there was a linear relationship between the CV(image) and CV(Bn) (.) For example, to obtain a CV(Bn) = 0.10 in an image with CV(image) = 0.60, a sampling area of 400 x 400 pixels (11% of whole image) was required, whereas if CV(image) = 1.46, a sampling area of 1000 x 100 pixels (69% of whole image) became necessary. This suggests that a large-size dispersion of element sizes in an image requires increasingly larger sampling areas or a larger number of images.

Publication types

  • Research Support, Non-U.S. Gov't
  • Review

MeSH terms

  • Computer Simulation
  • Food Analysis / methods*
  • Fruit / chemistry
  • Fruit / cytology
  • Image Processing, Computer-Assisted / methods*
  • Malus / chemistry
  • Malus / cytology
  • Models, Statistical
  • Statistics as Topic