J Chromatogr A, 1628, 461461 (2020). Samples were hydro-methanolic extracts of 100 genuine saffron samples (Crocus sativus stigmata, Iridaceae) from South Khorasan (SK) and Razavi Khorasan (RK) provinces (Iran), pure or mixed in several proportions with common vegetal adulterants: C. sativus style, Calendula officinalis petals (Asteraceae, Asteroideae), Carthamus tinctorius petals (Asteraceae, Carduoideae), Rubia tinctorum rhizomes (Rubiaceae). Commercial saffron samples (containing artificial adulterants) were also tested. TLC on silica gel with ethyl acetate – methanol – water – acetic acid 66:23:11:1. Evaluation at 254 nm, 366 nm, and 440 nm. Crocin (carotenoid, hRF 38) was used for optimization of extraction (parameters being first calculated by chemometry), using multilinear regression and ANOVA. Image data (pixel intensities and colors of each sample under the three selected wavelengths) were unfolded into a data matrix and transformed into a vector, used for multivariate image analysis of the chromatogram fingerprints. This allowed: A) separation of genuine samples by principal component analysis (PCA) into 2 clusters according to origin (cold climate in Northern half of RK vs. warm climate in SK and Southern part of RK) with 92 % prediction accuracy; B) separation of samples according to purity / vegetal adulterant groups by partial least squares – discriminant analysis (PLS-DA) with 98 % accuracy (if 10 µL extract applied); C) separation with 100 % prediction accuracy by PCA between genuine, mixed, and commercial samples.

J. Liq. Chromatogr. Relat. Technol. https://doi.org/10.1080/10826076.2021.1932521 (2021). HPTLC of gallic acid in the stem bark of Schinus terebinthifolius on silica gel with toluene - ethyl acetate - formic acid - methanol 15:15:4:1. The plates were scanned at 254 nm and 366 nm. The hRF value for gallic acid was 43. Image features were acquired using a combination of two approaches: Haralick texture features and Zernike moments. The GNU OctaveVR software was used to set the architectures of the Artificial Neural Network. The mathematical data provided by the image analysis was correlated with the gallic acid content determined by HPLC. The method allowed the prediction of phenolic content through TLC plate images.

Chinese J. Nucl.Tech.(Hejishu) 10 (12), 24-28 (1987) . Description of a pneumatic position-sensitive proportional counter as a radio detector for TLC, consisting of a counter, a time-amplitude converter and a computerized data system for detection of radioactive element labeled compounds.

Chromatographia 27, 123-127 (1989). Study of overloaded systems using an equilibrium sandwich for continuous TLC with dyes as model samples. Investigation of the effect of sample volume and the mode of development on the maximum separation yield. Calculation of the final Rf values for the front and rear edges of the bands with a computer program. Good agreement between the experimental Rf values of zone boundaries and the values predicted by computer simulation.

J. Planar Chromatogr. 2, 71-75 (1989). A validation procedure for TLC scanners is presented including determination of signal to noise ratios and instrumental sensitivity. Using test plates, the overall reliability can be determined. Validation of individual parts of the software is done by comparison of different algorithms, for example smoothing algorithms. Validation of spotting devices is shown to be possible by analysis of variances from different types of scans in order to separate errors created by positioning and measurement.

Phytochem. Anal. 20, 19-23 (2009). TLC of curcumin (1), demethoxycurcumin (2), and bisdemethoxycurcumin (3) from the rhizomes of Curcuma longa on silica gel with hexane – chloroform – methanol 10:10:1 as mobile phase. Quantitative determination by recording the chromatogram using a digital scanner and analyzing the density of the TLC spot with the Scion Image software. The hRf values of (1), (2), and (3) were 42, 25, and 18, respectively. Selectivity regarding matrix was given. Linearity was between 0.375 and 6 µg/spot for all curcuminoids. The intermediate precision of the method was satisfactory. Recovery was 101.9 % for (1), 104.8 % for (2), and 101.5 % for (3). The limits of detection and quantification were 43 and 143 ng/spot for (1), 69 and 230 ng/spot for (2), and 73 and 242 ng/spot for (3). The method was compared with an official densitometric method, and the analytical results were not significantly different.

J. Planar Chromatogr. 2, 138-141 (1989). To overcome structure signals of the HPTLC-plates the concept of multiple scanning at changed angles of a planar chromatogram rotated stepwise was used. The procedure does not falsify raw data and has nothing to do with any kind of smoothing. The authors work with the mean values but do not touch the original raw data.

Anal. Chim. Acta 641 (1-2), 52-58 (2009). Comparison of classical filtering techniques (Savitzky–Golay, Adaptive Degree Polynomial Filter, Fourier denoising, Butterworth and Chebyshev IIR filters) and wavelet shrinkage (31 mother wavelets, 3 thresholding techniques and 11 decomposition levels) with the original noisy signal and a reference signal which was denoised experimentally by averaging 64 measurements. The best similarity to the reference signal was obtained with filters, however the signal was slightly oversmoothed. The wavelet shrinkage method gave less denoised signals. There was a significant influence of the thresholding technique and decomposition level, and best conditions were at level 2 or 3 and soft thresholding), whereas changing of the mother wavelet almost did not change the result. The presented results can be used as general recommendations for denoising densitometric fingerprints before applying further chemometric algorithms. The best choices were: Savitzky–Golay filter of appropriate window width (optimized against autocorrelation) or wavelet shrinkage with Haar wavelet, soft thresholding and high decomposition level.