Fig 1.
(a) VideoMeterLab, (b) acquired data cube; here presenting a minced meat sample.
Fig 2.
Overview of the automated segmentation.
(a) Original image input, preprocessing and normalization, (b) EBMC based wavelength band selection, (c) mixture modeling for threshold definition for background removal, (d) binary mask of meat product sample, (e) EBMC based wavelength band selection for fat and connective tissue detection using (f) mixture modeling for threshold definition, (g) binary mask of final informative area and (h) application of the final binary mask to all 18 spectral bands.
Fig 3.
Linear regression of the area sizes (percentages, i.e. ratio of informative detected areas to the total area of the image) resulted by the two methods VM (VideoMeterLab) and AS (Automated Segmentation).
Table 1.
α, parameter α of the linear regression model (f(x) = ax + β); β parameter β of the linear regression model; R-square, goodness of fit; RMSE, root mean square error; r, correlation coefficient; p-value, the p-value of the hypothesis that correlation between values are random.
Fig 4.
Analysis of mean reflectance values of the detected informative area at each wavelength and their corresponding standard deviations for minced meat dataset.
It is obvious that all values are almost identical, i.e. we get the same information by either method, AS or VM.
Fig 5.
(a) mean colocalized areas for the three datasets on the basis of VM’s detected informative area; error bars represent the standard error of the colocalization, (b) “Anti-colocalization” percentages means and standard error (anti-colocalization stands for areas that have not been detected by the other method).
Table 2.
mean colocalization; Standard Error (SE);