Classification images have recently become a widely used tool in visual psychophysics. Here, I review the development of classification image methods over the past fifteen years. I provide some historical background, describing how classification images and related methods grew out of established statistical and mathematical frameworks and became common tools for studying biological systems. I describe key developments in classification image methods: use of optimal weighted sums based on the linear observer model, formulation of classification images in terms of the generalized linear model, development of statistical tests, use of priors to reduce dimensionality, methods for experiments with more than two response alternatives, a variant using multiplicative noise, and related methods for examining nonlinearities in visual processing, including second-order Volterra kernels and principal component analysis. I conclude with a selective review of how classification image methods have led to substantive findings in three representative areas of vision research, namely, spatial vision, perceptual organization, and visual search.

Realization of graphene-based sensors and electronic devices remains challenging, in part due to integration challenges with current fabrication and manufacturing processes. Thus, scalable methods for in situ fabrication of high-quality graphene-like materials are essential. Low-cost CO₂ laser engravers can be used for site-selective conversion of polyimide under ambient conditions to create 3-D, rotationally disordered, few-layer, porous, graphene-like electrodes. However, the influences of non-linear parameter terms and interactions between key parameters on the graphitization process present challenges for rapid, resource-efficient optimization. An iterative optimization strategy was developed to identify promising regions in parameter space for two key parameters, laser power and scan speed, with the goal of optimizing electrode performance while maximizing scan speed and hence fabrication throughput. The strategy employed iterations of Design of Experiments Response Surface (DoE-RS) methods combined with choices of readily measurable parameters to minimize measurement resources and time. The initial DoE-RS experiment set employed visual response parameters, while subsequent iterations used sheet resistance as the optimization parameter. The final model clearly demonstrates that laser graphitization through raster scanning is a highly non-linear process requiring polynomial terms in scan speed and laser power up to fifth order. Two regions of interest in parameter space were identified using this strategy: Region 1 represents the global minimum for sheet resistance for this laser (∼16 Ω/sq), found at a low scan speed (70 mm/s) and a low average power (2.1 W) . Region 2 is a local minimum for sheet resistance (36 Ω/sq), found at higher values for scan speed (340 mm/s) and average power (3.4 W), allowing ∼5-fold reduction in write time. Importantly, these minima do not correspond to constant ratios of average laser power to scan speed. This highlights the benefits of DoE-RS methods in rapid identification of optimum parameter combinations that would be difficult to discover using traditional one-factor-at-a-time optimization. Verification data from Raman spectroscopy showed sharp 2D peaks with mean full-width-at-half-maximum intensity values <80 cm^–1 for both regions, consistent with high-quality 3D graphene-like carbon. Graphene-based electrodes fabricated using the parameters from the respective regions yielded similar performance when employed as capacitive humidity sensors with hygroscopic dielectric layers. Devices fabricated using Region 1 parameters (16 Ω/sq) yielded capacitance responses of 0.78 ± 0.04 pF at 0% relative humidity (RH), increasing to 31 ± 7 pF at 85.1% RH. Region 2 devices (36 Ω/sq) showed comparable responses (0.88 ± 0.04 pF at 0% RH, 28 ± 5 pF at 85.1% RH).

Every biological or artificial visual system faces the problem that images are highly ambiguous, in the sense that every image depicts an infinite number of possible 3D arrangements of shapes, surface colors, and light sources. When estimating 3D shape from shading, the human visual system partly resolves this ambiguity by relying on the light-from-above prior, an assumption that light comes from overhead. However, light comes from overhead only on average, and most images contain visual information that contradicts the light-from-above prior, such as shadows indicating oblique lighting. How does the human visual system perceive 3D shape when there are contradictions between what it assumes and what it sees? Here we show that the visual system combines the light-from-above prior with visual lighting cues using an efficient statistical strategy that assigns a weight to the prior and to the cues and finds a maximum-likelihood lighting direction estimate that is a compromise between the two. The prior receives surprisingly little weight and can be overridden by lighting cues that are barely perceptible. Thus, the light-from-above prior plays a much more limited role in shape perception than previously thought, and instead human vision relies heavily on lighting cues to recover 3D shape. These findings also support the notion that the visual system efficiently integrates priors with cues to solve the difficult problem of recovering 3D shape from 2D images.