Usability of Digital Cameras for Verifying Physically Based ...

More documents

Recommendations

Info

Eighteen observers were asked to match the gray levels of the presented setting to a predefined set of samples, i.e. to judge the lightness of the objects. For this purpose, they have been trained to do this by recalling the different gray levels from memory before the actual experiment. The 11 different settings – the 10 images and the real scene – were presented in random order. Afterwards, the gray levels chosen by each participant in an image were compared with the values chosen in the real scene. The closer these values were, the closer the image was considered to be to the real scene. In summary, the results of this study showed that high quality Radiance images (see figure 4(d) and (e)), the tone-mapped image (see figure 4(g)) and even one of the defective images (see figure 4(h)) were good representations of the actual scene. Though, the low quality Radiance simulations (see figure 4(b) and (c)), the estimated materials image (see figure 4(f)), the raytraced image (see figure 4(i)) and the radiosity image (see figure 4(j)) differ considerably from the real setting. 2.2 Experimental Measurements 2.2.1 Improved Cornell Box One year after the first experiments in Cornell (see section 2.1.1), the radiosity algorithm was improved by projecting onto an imaginary cube instead of a sphere. This hemi-cube radiosity [CG85] was again verified using the Cornell Box approach. Now, radiometric measurements of the cube were taken and compared with the results of the algorithm. In the context of the development of the hemi-cube radiosity algorithm, the importance of verifying the implementation of a rendering algorithm was recog- nized. In 1986, Meyer et al. [MRC + 86] investigated the procedure of experimental evaluation of rendering systems in detail. The following quotation em- phasizes the necessity of experimental verification: “If a scientific basis for the generation of images is to be established, it is necessary to conduct experimental verification on both the com- ponent steps and the final simulation.” [MRC + 86] The assembly of the Cornell Box had slightly changed since the first experiments had been made. The light bulbs and the diffuse white surfaces were replaced 10
y a light source on top of the cube. A small rectangular opening was cut into the top panel and covered with a piece of opal glass to provide diffuse light. An in- candescent flood light was mounted 15 inches above on top of a metal cone whose interior was painted white. In order to avoid interreflections with the surrounding, the box was placed on a black table and the walls were covered with black fabric. Moreover, the panels of the box could be exchanged by other panels with different colors. In order to render an image that can be compared to the real world scene, the properties of the light source and the surfaces have to be measured. The spectral energy distribution of the light that shone through the opal glass was acquired by a method described by Imhoff [Imh83]. Moreover, the intensity and the direction of the light were needed for the calculations. They were measured by using a pho- tometer (see appendix B.2) combined with an infrared filter. The inaccuracies of the device have been adjusted by a correction factor. Another piece of equipment – a spectrophotometer – was employed to measure the spectral reflectances of the surfaces that were painted in different colors. For radiometric comparisons, a device is needed that accurately simulates the behavior of the human eye, like the ability to capture multiple samples at each wavelength band of light. Since such a device was not available, Meyer et al. obtained a radiometer (see appendix B.1) instead. This device was able to measure over the range of the radiometric spectrum but provided just a single reading, which is not enough to represent a whole scene. Therefore, 25 evenly distributed measurements were taken (see figure 5). The sample points were chosen that way to avoid shadows and to maximize the amount of light that hits the probe. Three different test scenes were created to analyze the verification procedure. Two of them consisted of an empty cube – the first had only white panels, the second contained one blue panel. The third test scene was a white cube with a white box inside it. Figure 6 shows the results of the third test scene. The cube is tilted in a way that the front panel is on top and the top panel is facing the left side. The irradiation H is shown on the vertical axis. The red lines show the result calculated by the radiosity algorithm whereas the blue lines represent the actual measurements. The correspondence between the values of the computer generated image and the real scene is clearly visible. They have a root mean square difference of about four percent. It has to be mentioned that the described method only works with diffuse envir- 11
Page 1 and 2: DISSERTATION Usability of Digital C
Page 3 and 4: Kurzfassung Physically Based Render
Page 5 and 6: Acknowledgements I want to thank We
Page 7 and 8: 3 Colorimetry 32 3.1 Color Spaces .
Page 9 and 10: E.5 Treating the Camera as a Black
Page 11 and 12: 20 The CIE 1931 chromaticity diagra
Page 13 and 14: 53 (a) Density plot for using the c
Page 15 and 16: List of Tables 1 Camera settings th
Page 17 and 18: 1 Introduction In recent years, a l
Page 19 and 20: 1.3 Applications of Physically Base
Page 21 and 22: matter are calculated in XYZ space.
Page 23 and 24: Figure 2: A photograph of the real
Page 25: (a) Photo (b) Default (c) 2 Bounces
Page 29 and 30: Figure 7: Points A to F were select
Page 31 and 32: Figure 9: A difference image of pho
Page 33 and 34: Figure 12: From top to bottom: a hi
Page 35 and 36: (a) Figure 13: (a) Left side: measu
Page 37 and 38: Figure 14: Six photocells were posi
Page 39 and 40: Figure 16: The left image shows a r
Page 41 and 42: (a) (b) Figure 18: (a) The validati
Page 43 and 44: shape of the caustic. 2.3 Analytica
Page 45 and 46: two test cases, but instead of the
Page 47 and 48: segments of the sphere and its actu
Page 49 and 50: of luminance [GJB80]. In XYZ color
Page 51 and 52: where L ∗ � � Y m = 903.3 Yn
Page 53 and 54: (a) (b) Figure 21: Chromaticity dia
Page 55 and 56: 4 Digital Cameras Pattanaik et al.
Page 57 and 58: Figure 23: The RAW image data is co
Page 59 and 60: 5 Physically Based Rendering System
Page 61 and 62: see section 7.4. By avoiding XYZ sp
Page 63 and 64: Figure 27: The sensor plane marking
Page 65 and 66: (a) (b) Figure 30: (a) ColorChecker
Page 67 and 68: available on MAC systems. We chose
Page 69 and 70: (a) (b) (c) Figure 34: Standard dev
Page 71 and 72: Figure 35: Eighty measurement spots
Page 73 and 74: (a) (b) (c) (d) Figure 37: (a) Lumi
Page 75 and 76: (a) (b) Figure 40: (a) Luminance di
Page 77 and 78:
i \ j 0 1 2 3 4 5 6 7 8 9 10 11 12
Page 79 and 80:
(a) (b) (c) (d) Figure 43: (a) Lumi
Page 81 and 82:
(a) (b) Figure 46: Standard deviati
Page 83 and 84:
(a) (b) Figure 49: (a) The measurem
Page 85 and 86:
of each patch using a spectrophotom
Page 87 and 88:
Figure 51: L ∗ a ∗ b ∗ differ
Page 89 and 90:
(a) (b) Figure 53: (a) Density plot
Page 91 and 92:
(a) (b) (c) Figure 55: Histograms o
Page 93 and 94:
Using an ICC profile obviously impr
Page 95 and 96:
(a) (b) (c) (d) Figure 58: Histogra
Page 97 and 98:
5 6 7 8 9 10 11 Min 0.50 0.49 0.13
Page 99 and 100:
in the light transport simulation o
Page 101 and 102:
the internal data of a camera and w
Page 103 and 104:
etter results: One can measure the
Page 105 and 106:
Figure 62: A Mathematica generated
Page 107 and 108:
their internal color space. This mi
Page 109 and 110:
Figure 66: Surface reflectance valu
Page 111 and 112:
as bad compared to other methods. S
Page 113 and 114:
in sufficient detail. The approach
Page 115 and 116:
A Radiometric and Photometric Quant
Page 117 and 118:
B.6 Goniophotometer A goniophotomet
Page 119 and 120:
C.3 Color Chart Figure 68: A Konica
Page 121 and 122:
D Web Resources of Validation Data
Page 123 and 124:
8 bit jpg 16 bit 29 11.58 11.52 11.
Page 125 and 126:
8 bit jpg 16 bit 91 17.21 17.11 17.
Page 127 and 128:
E.2 Using an ICC Profile 8 bit jpg
Page 129 and 130:
8 bit jpg 16 bit 62 16.10 16.08 16.
Page 131 and 132:
8 bit jpg 16 bit 124 11.08 10.89 11
Page 133 and 134:
1 2 3 4 31 9.10 3.61 4.26 5.96 32 8
Page 135 and 136:
1 2 3 4 93 6.25 4.11 4.05 8.69 94 4
Page 137 and 138:
E.4 Adaptive Mapping - Higher Order
Page 139 and 140:
5 6 7 8 9 10 11 62 3.42 2.52 2.51 2
Page 141 and 142:
5 6 7 8 9 10 11 124 2.10 2.14 2.96
Page 143 and 144:
R G B 31 -38.94 -25.49 29.26 32 -9.
Page 145 and 146:
R G B 93 103.86 109.28 55.06 94 31.
Page 147 and 148:
References [3dS06] 3D Studio Max: w
Page 149 and 150:
[HLR00] Guowei Hong, M. Ronnier Luo
Page 151 and 152:
Color Imaging Conference, pages 250
Page 153:
[VMKK00] V. Volevich, K. Myszkowski
show all

Usability of Digital Cameras for Verifying Physically Based ...

Create successful ePaper yourself

Delete template?

Save as template?