Goal and Motivation

Goal and Motivation

Goal and Motivation

• To study our (in)ability to detect inconsistencies in the illumination of objects in images

• Invited Talk!– Hany Farid: Photo Forensincs: Lighting and Shadows

Goal and Motivation

Goal and Motivation

• Suggest thresholds for error limits in image-based light detection algorithms

• Underconstrained pb.

Previous work

• Todd and Mingolla [1983] low accuracy of HVS using lightprobes to infer light direction

• [Mingolla and Todd 1986] HVS does not assume objects as diffuse by default.

• Koenderik et al. [2004] HVS increases accuracy detecting the light field direction when shadow boundaries are present.

Previous work

• [Ostrovsky et al. 2005] HVS can easily spot an anomalously lit object in an array of identical objects with the same orientation and lit exactly the same.

• O’Shea et al. [2008] for unknown geometries the angle between the viewing direction and the light direction is assumed to be 20-30 degrees above the viewpoint.

• Did I mention the invited talk already?

Overview

• Experiment #1 The goal is to suggest a general threshold for diffuse and shiny objects under different light configurations.

Overview

• Experiment #1 The goal is to suggest a general threshold for diffuse and shiny objects under different light configurations.

• Experiment #2 Analysis of the influence of texture properties (spatial frequency) in the perception process.

Overview

• Experiment #1 The goal is to suggest a general threshold for diffuse and shiny objects under different light configurations.

• Experiment #2 Analysis of the influence of texture properties (spatial frequency) in the perception process.

• Experiments #3 and #4 Designed to explore how well our findings carry over to real images. Experiments with modified photographs as stimulus.

Experiment #1

• A series of images were shown. All with several objects lit (directional lighting) from the same angle… except for one

• Select the inconsistently lit object in each image

• The images were randomly presented• Only vary the more restrictive slant angle [Koenderink

04]

Experiment #1

• Example of image used in the test

Experiment #1

• This experiment had 3 dimensions:– Angle of divergence: 0-90 degrees, in 10-degree increments– Spatial configuration of lights : both in the front, both in the

back, mixed– Shininess property: Highlights - NO Highlights

Experiment #1

• In total 10x2x3 = 60 images were generated

• 55 participants took the test: ages 16-58, 33 male, 22 female. 18 had an artistic background.

Experiment #1: Results

• Up to 20 degrees of divergence the probability of detection is around chance (12:5%).

• If both lights are in the front: up to 30 degrees– agree with [Koenderink et al. 2004] which suggested that

shaded areas and self-shadows increase our accuracy.

Experiment #1: Results

• The performance of HVS is slightly lower when highlights are present Todd and Mingolla’s [1983]

• Diverges from some computer vision approaches which do use highlights as visual cues [Lagger and Fua 2006].

Experiment #2

• We aim to analyze the influence in the perception process of the spatial frequency of the texture.

• The psychophysical test consists of a new series of images, which has been shown to 32 users (ages 22-57; 23 male and 9 female).

• The test was displayed using the same methodology as in Experiment One.

Experiment #2

• Example of image from the test. Four textures with different spatial frequency x 10 divergence degrees = 40 images shown to each user.

Experiment #2: Results

• Responses provided by users in the test, shown by texture frequency.

Experiment #2: Results

• Higher frequencies do mask lighting inaccuracies up to the detection threshold of 20-30 degrees, making the detection task more difficult.

• For angles > 40 degrees we found no significant difference (p > 0:05) in the results the visual system may not take intensity variations due to the surface material as suggested in [Khang et al. 2006]

Experiment #3

• This test consists of a simple scene containing a set of eight real objects

• The scene was photographed three times: the original scene, plus two more with the angle of the main light source varying 20 and 30 degrees respectively.

• Two images were obtained by compositing the original image with a pair of objects (ceramic purple doll and the Venus figurine) from the two images with varying light sources.

Experiment #3

Experiment #3

Experiment #3

Experiment #3

• 25 users (ages 17-62, 14 male and 11 female)• Each user was shown one image with two

inconsistently lit objects (both 20 or 30 degrees).• They were asked the following question:

In the following image one or two objects have been inserted and they have a different illumination than the rest of the scene. Could you point it/them out?

Experiment #3

• These results motivate the test #4.

• Hit ratio is below chance for one (40,625%) and two objects (3,125%) with both 20 and 30 degrees of divergence.

Experiment #4

• This test is designed to narrow the threshold range anticipated in tests #1 and #3 for real images.

• Nine versions of a new scene were generated.• Four photographs of the same scene were taken at 0,

20, 30 and 40 degrees of divergence from a reference direction.

• Three objects were masked out and only one object was combined at a time 3 objects x 3 directions = 9 images.

Experiment #4

• The objects with light modified illumination.

Experiment #4

• 60 users (ages 18-59, 38 male and 22 female)

• Each user was shown three images with a random inconsistently lit object at 20, 30 and 40 degrees of divergence.

• The same object was never shown more than once per user.

Experiment #4

• The results show a trend similar to the tests with synthetic objects

• However the thresholds are more conservative (30-40 degrees instead of 20-30)

• Reasons richer visual cues? Naturalness of the scene?

Conclusions

• We have presented four different tests to measure the accuracy of human vision detecting lighting inconsistencies in images.

• The results of our experiments agree with previous research [Ostrovsky et al. 2005; Koenderink et al. 2004; Lopez-Moreno et al. 2009].

• We suggest a perceptual threshold for multiple configurations: materials, position of light sources,… .

Acknowledgments

• This research was partially funded by a generous gift from Adobe Systems Inc, the Gobierno de Aragόn (projects OTRI 2009/0411 and CTPP05/09) and the Spanish Ministry of Science and Technology (TIN2007-63025).