Commentaries on Stanley Klein's Research Articles
A. Applied Vision
82. Silverstein, D.A. & Klein, S.A. (1994). Restoration of compressed images. Image and Video Compression Majid Rabbani and Robert J. Safranek, Editors, Proc. SPIE 2186, 56-64.
109. Silverstein, D.A. & Klein, S.A. (1996). Precomputing and encoding compressed image enhancement instructions. Submitted to IEEE Image Processing.
These two papers developed a scheme to improve the quality of compressed images without a significant increase in the number of transmitted bits. The core idea is that during compression one still has access to the original scene or image sequence so several enhancement schemes for restoring the compressed image can be tested and compared. Our trick lies in how to hide the enhancement information. Manuscript #109 calculates the very minor increase in bits for a very large improvement in fidelity. We have applied for a patent for these ideas with the support of the UC Technology Transfer office.
86. Klein, S.A. & Silverstein, D.A. (1994). Image compression, fidelity metrics and models of human observer. Proc. of the International Picture Coding Symposium, PCS'94. 96 - 99.
This paper at the International Picture Coding annual meeting presented a brief summary of our groups' activity in the realm of image compression. We describe several areas where a good model of the human visual system could be used to improve image compression algorithms. As a simple example we present the Fourier amplitudes of the DCT basis functions (derived by us previously) and discuss the connection to the contrast sensitivity function. We also summarize our image enhancement schemes (articles #82 and #109).
102. Carney, T., Klein, S.A. & Hu, Q.J. (1996). Visual masking near spatiotemporal edges. Human Vision and Electronic Imaging. BE Rogowitz & JP Allebach, Eds, Proc. SPIE 2657, 2657-45.
This is the first of our articles based on data from Qingmin Hu's PhD thesis. He measured the visibility of a thin line probe, flashed for one frame in the presence of a masking bar of variable width that was flashed on for 500 msec. The time of the probe relative to the bar onset was varied over a wide range, with most data gathered for asynchronies between -50 and + 50 msec. This is a combination of Crawford masking and the Westheimer effect. Of special interest is that we used all possible polarities of probe and mask, including cases where only the periphery contained the mask and the probe region was static. The resulting data is quite rich with some surprising polarity specific effects whereby the peak masking threshold elevation depends on the relative polarity and timing of test and mask. We are pursuing these effects both because we have found that they are difficult to predict and because they may be important for compression of image sequences.
83. Hu, Q.J. & Klein, S.A.(1994). Correcting the adjacent pixel nonlinearity on video monitors. Human Vision, Visual Processing and Digital Display V BE Rogowitz & JP Allebach, Eds, Proc.
SPIE 2179, 37-46.
91. Klein, S.A., Carney, T. & Hu, Q.(1995) Improved lookup table to correct CRT pixel nonlinearity. Human Vision, Visual Processing and Digital Display VI BE Rogowitz & JP Allebach, Eds, Proc. SPIE 2411, 170 - 179.
104. Klein, S.A., Hu, Q.J. & Carney, T.(1996). The adjacent pixel nonlinearity: Problems and solutions. Vision Res., 36, 3167-3181.
- close section -
103. Klein, S.A. & Carney, T.(1996). Transducer function for energy (viewprint) models. Society for Informational Display 96 Digest 27, 425-428.
I was shocked to discover that we didn't know the precise shape of the ideal observer's transducer function for detecting a sinusoidal granting when the phase was randomized from trial to trail. Although this situation is briefly discussed by Green and Swets we weren't aware of anyone actually calculating the transducer shape. We carried out Monte Carlo simulations for the ideal observer with phase uncertainty and determined the shape of the transducer function. This calculation is quite relevant to the issue of masking in real images since it is often the case the phase of the error signal is not known. Thus the d' reduction produced by phase uncertainty should be incorporated into general fidelity metrics. In addition we discussed energy (complex cell) representations where phase is discarded vs. more standard filter or Fourier (simple cell) representations where phase is preserved. We pointed out the usefulness of the energy approach to fidelity metrics.
- close section -
B. Understanding vision in the periphery and in amblyopia (expand)
87. Levi, D.M., Klein, S.A. & Wang, H.(1994a). Amblyopic and peripheral Vernier acuity: A test-pedestal approach. Vision Research 34, 3265 - 3292.
88. Levi, D.M., Klein, S.A. & Wang, H.(1994b). Discrimination of position and contrast in amblyopic and peripheral vision. Vision Research 34, 3293 - 3314.
This pair of papers represents our application of the test-pedestal approach to clarifying the amblyopic deficit. #87 uses multipole stimuli (edge, line, dipole and quadrupole) and #88 uses sinusoidal stimuli. These papers provide the cleanest evidence for the differences between strabismic and anisometropic amblyopia. We find that the anisometropic loss in vernier acuity can be fully accounted for by the loss in visibility such as would be expected from loss of high spatial frequency mechanisms and/or problems with contrast processing. For strabismic amblyopes we find an extra loss on the vernier position task beyond what can be explained by a degraded contrast sensitivity function. We have made this point in previous papers, but in this pair of papers based on the test-pedestal approach the data make the case without requiring additional assumptions.
101. Levi, D.M. and Klein, S.A.(1996). Limitations on position coding imposed by undersampling and univariance. Vision Research, 36, 2111-2120.
This paper is part of a long term debate with Robert Hess on the nature of the amblyopic loss. The present topic is concerned with whether the loss found in strabismic amblyopia might be due to undersampling. Hess believes the loss is due to scrambling of connections rather than undersampling. Dr. Levi and I believe that both factors contribute. In a 1993 Vision Research article Hess and Field claimed that the principle of univariance eliminates the possibility that undersampling might be responsible. I article #101 we present several arguments, several simulations and some data that point out a number of fallacies in the Hess & Field position. The new data discussed in the next article (#108) provides further evidence for our position.
108. Wang, H., Levi, D.M. & Klein, S.A.(1996). Spatial uncertainty and sampling efficiency in amblyopic position acuity. Submitted to Vision Research.
About 5 months ago we submitted a paper to Vision Research that measured the effective number of samples available to anisometropic and strabismic amblyopes. We used the methodology described in article #100. We found that both anisometropic and strabismic amblyopes had increased levels of intrinsic uncertainty, si but only strabismic amblyopes had a dramatic loss of effective number of samples. This technique provides a powerful way of discriminating between the different factors contributing to the amblyopic loss. We are sorry we didn't have this data available when we wrote article #101.
- close section -
C. Evoked Potentials (expand)
84. Baseler, H.A., Sutter, E.E., Klein, S.A. & Carney, T.(1994). The topography of visual evoked response properties across the visual field. Electroenceph. &
Clinical Neurophys. 90, 65-81.
96. Klein, S.A. & Carney, T.(1995).The usefulness of the Laplacian in principal component analysis and dipole source localization. Brain Topography, 8, 91 - 108.
These two articles lay the groundwork for my most ambitious research project: the localization of sources of the visual evoked potential (VEP). Over the past forty years there have been numerous attempts to solve this problem, but once more than two sources are present the source locations become unreliable. Interest in this area has received a tremendous boost because functional MRI and the growth of cognitive neuroscience has increased attention to this field. The VEP has a thousand-fold better temporal resolution than f-MRI and we believe it also has much better signal to noise properties. We have developed a number of new approaches, detailed in articles #84 and #96, that make us optimistic that we can succeed where previous investigators have failed. Our most important innovation is to combine the standard multielectrode technique (my lab has a 64 channel setup) with Erich Sutter's multiple stimulus method: 56 small visual field areas are simultaneously recorded. The continuity of adjacent retinotopic areas is used to disambiguate the source solutions. This project is Scott Slotnick's PhD thesis area. In article #96, we demonstrate how the source localization can be substantially improved by having the search algorithm work on the Laplacian of the electrode voltages rather than on the voltages themselves.
134. Slotnick, S.D., Klein, S.A., Carney, T., Sutter, E.E. & Dastmalchi, S.D.(1999). Using multi-stimulus VEP source localization to obtain a retinotopic map of human primary visual cortex. Clinical Neurophysiology. 110, 1793-1800.
Slotnick S.D., Klein S.A., Carney, T. & Sutter, E.(1999) A direct method for estimating the visual space scaling factor of the human visual cortex. Submitted to Nature.
These two articles are part of Scott Slotnick's PhD Dissertation and represent our first step in implementing our new method of isolating the VEP sources. The crowded, rapidly flickering, high spatial frequency stimulus that we used preferentially excites primary visual cortex, V1. We fit the data with a single dipole source (one source for each of the 60 stimulus patches) and was able to account for more than 50% of the variance (noise included). The first article goes into all the methodological details, the most important one being our use of a common time function for nearby sources. The article demonstrate coherent motion of the dipoles as expected from the topography of V1. The second article uses the source localization data to estimate the cortical magnification factor (mm of cortex per degrees of visual field). Two methods were used to estimate magnification. Both estimates are in good agreement with anatomy and psychophysics. This is the first, accurate, direct estimate of the magnification factor in normal individuals.
- close section -