Timo Jaaskelainen

Unsupervised filtering of color spectra

We describe a class of unsupervised systems that extract features from databases of reflectance spectra that sample color space in a way that reflects the properties of human color perception. The systems find the internal weight coefficients by optimizing an energy function. We describe several energy functions based on second- and fourth-order statistical moments of the computed output values. We also investigate the effects of imposing boundary conditions on the filter coefficients and the performance of the resulting systems for the databases with the reflectance spectra. The experiments show that the weight matrix for one of the systems is very similar to the eigenvector system, whereas the second type of system tries to rotate the eigenvector system in such a way that the resulting filters partition the spectrum into different bands. We also show how the system can be forced to use weight vectors with positive coefficients. Systems consisting of positive weight vectors are then approximated with Gaussian quadrature methods. In the experimental part of the paper we investigate the properties of three databases consisting of reflectance spectra. We compare the statistical structure of the different databases and investigate how these systems can be used to explore the structure of the space of reflectance spectra.

Spectral representation of color images

An approach to color image representation is described that allows high accuracy in the spectral dimension and that leads to possibilities in color image analysis and to novel analysis methods. A color image can be stored in the same space in the computers memory as a conventional red, green, and blue image, which makes it possible to reconstruct the whole color spectrum of each pixel. Fast optically realizable parallel processing implementations are also discussed.<<ETX>>

The photochromic properties of 4-keto bacteriorhodopsin

The all-trans-retinal of the native chromoprotein of bacteriorhodopsin (BR) purple membrane was replaced with the synthetic 4-keto retinal. Gelatin films were made from the reconstituted BR and the intensity-dependent basic optical properties were investigated. It is shown that at room temperature the M-state is composed of two different absorption bands (maximum at 413 nm and 435 nm) with different relaxation times. Unexpected low-intensity transmission properties which cannot be explained by the two-level model are reported. The M-state of the 4-keto BR is shown to be about 100 times slower than the M-state of the similar wild type BR-film.

Eigenvector interpretation of the Farnsworth–Munsell 100-hue test

We measured the reflectance spectra for the 85 color caps of the Farnsworth-Munsell 100-hue test. Eigenvectors and eigenvalues of a correlation matrix of cone responses were computed, with the cone responses being determined from the 85 test caps, arranged in order (according to color) by means of a linear model. It is shown that the Farnsworth-Munsell 100-hue test can be simulated by use of eigenvectors of the cone responses. The eigenvectors can be interpreted as nonopponent signal and opponent color signals. The normal observer can determine the color of a cap by using two opponent color signals. For color-blind persons (dichromats) one or the other opponent signal is defective, and errors can occur during the test. The simulation results also suggest that eigenvectors can be used to predict results of arrangement tests similar to the Farnsworth-Munsell 100-hue test.

Co-occurrence matrix and self-organizing map-based query from spectral image database

A co-occurrence matrix and Self-Organizing Map (SOM) based technique for searching images from a spectral image database is proposed. At first the SOM is trained and the Best Matching Unit (BMU) histogram is created for every spectral image of a database. Next, the texture-histogram is calculated from the co-occurrence matrices, generated using the 1st inner product images of the spectral images. BMU-histogram and the texture-histogram are combined to one feature histogram and these histograms, generated for each spectral image of a database, are saved to a histogram database. The dissimilarities between the histogram of the query image and the histograms of the database are calculated using different distance measures, more precisely Euclidean distance, dynamic partial distance and Jeffrey divergence. Finally, the images are ordered according to the histogram dissimilarity. The results using a real spectral image database are given.

Nonsinusoidal absorption gratings in photographic materials

Abstract The diffraction efficiencies of pure absorption gratings are investigated in photographic materials using a coupled multiwave analysis with a realistic photographic film model. It is, for example, shown that the maximum efficiency of the order +1 of non-sinosoidal absorption gratings is almost independent of the generally used regime indicator Q ′. Examples of diffraction characteristics for thin absorption gratings are included.

Wavelength interval in accurate tristimulus integration

Tristimulus integrals are approximated by summation equations. Now, interesting question is what should the wavelength interval be to obtain accurate tristimulus values? CIE recommends to do summation with 1 or 5 nm intervals. For dense intervals ASTM weighting sets are recommended to use. In that method weighting sets are calculated by Lagrange interpolation.