Lecture 17 Scanner Characterization and Calibration Sanjyot Gindi

Lecture 17 Scanner Characterization and Calibration - Sanjyot Gindi M. S. E. C. E, Purdue University July 18 th 2008 Papers/theses covering these materials can be found in the references under “Capture”

Objectives: n Spectral model based characterization of Samsung SCX 5530 scanner. n Empirical or regression based characterization of Samsung, HP Photosmart and Epson Photo RX 700 scanners. n Plotting of color gamuts and comparison of the 3 scanners based on them. Sanjyot Gindi Masters Thesis Presentation, July 18 th, 2008 Slide 2 of 43

Basic working of a flatbed scanner n ‘Target’ is placed on the glass top n A movable scanner-head consists of a lamp and sensors n Light from the lamp incident on the target is reflected back to the scan head. n This light passes through an optical Glass-top assembly and is received by sensors. n RGB values of the color of target – ‘device-dependent’ color space. Paper to be scanned Lamp Sanjyot Gindi Masters Thesis Presentation, July 18 th, 2008 Slide 3 of 43

There also Sheetfed Scanners n text Visioneer HP Laser. Jet Pro 400 Color MFP Sanjyot Gindi Masters Thesis Presentation, July 18 th, 2008 Slide 4 of 43

Modeling of a scanner system n Understanding the color characteristics and consistently predicting the color values. n [1] Two steps : q q Calibration: Linearization or gray balancing Characterization: Transformation from device dependent RGB values to co-ordinates in the device independent color space like CIE XYZ, L*a*b*. Sanjyot Gindi Masters Thesis Presentation, July 18 th, 2008 Slide 5 of 43

Methods of characterization 1. Regression based method: q Considers the scanner system as a ‘black box’ q Mapping from linear RGB to CIE XYZ 2. Model based method: q Uses known spectral functions of the components of a scanner system. Sanjyot Gindi Masters Thesis Presentation, July 18 th, 2008 Slide 6 of 43

1. Regression based method Rg Gg Bg n NLR NLG NLB Rl Gl X Transformation Bl Matrix T Y Z Rg, Gg , Bg are the values obtained from the scanner with gamma on, shading on. n n Rl, Gl, Bl are the linearized values X, Y, Z values are the CIE XYZ values obtained using the X-Rite Spectrophotometer with D 65 illuminant. n NL represents the non-linear relationship between output Rg, Gg , Bg values and Rl, Gl, Bl values Sanjyot Gindi Masters Thesis Presentation, July 18 th, 2008 Slide 7 of 43

Calibration procedure: n ‘Gray balancing’ using Y (Luminance) values of the neutral gray patches on the Kodak Q 60 target (see following slide). n Linear R, G, B values obtained using a power-law curve fit given by: , where R l = Y. Sanjyot Gindi Masters Thesis Presentation, July 18 th, 2008 Slide 8 of 43

Kodak Q 60 target Sanjyot Gindi Masters Thesis Presentation, July 18 th, 2008 Slide 9 of 43

Gray Balance curves a = 79. 63 b = 2. 272 X axis: R, G, B values from Scanner c = 0. 5905 (0 -255) Y axis: (Y) Luminance values from X-Rite (0 -100) a = 81. 58 b = 2. 033 c = 1. 344 Sanjyot Gindi a = 84. 14 b = 2. 198 c = 1. 093 Masters Thesis Presentation, July 18 th, 2008 Slide 10 of 43

Spectra-radiometers used for this study n text X-Rite DTP 70 Sanjyot Gindi Photo Research PR-705 Masters Thesis Presentation, July 18 th, 2008 Slide 11 of 43

Characterization: n Find scanner RGB values for Kodak Q 60 target : 240 color patches n CIE XYZ values of color patches using X-Rite spectrophotometer for D 65 illuminant condition. n Determine the 3 x 3 transformation matrix T given by: [ B ] = [ A ] [ T ] is found by least squares approximation where: 240 x 3 Sanjyot Gindi Masters Thesis Presentation, July 18 th, 2008 240 x 3 Slide 12 of 43

2. Model Based Method Lamp [L] R channel G channel B channel Target Kodak Q 60 reflectance [R] Sensor [F] Spectral Model of a Scanner Sanjyot Gindi Masters Thesis Presentation, July 18 th, 2008 Slide 13 of 43

Experiments: 1. The Lamp spectrum [L], is obtained using a spectroradiometer and a white diffusion target. 2. The Sensor spectral response [F], is obtained using the Monochromator (400 nm to 700 nm) 3. The Spectral Reflectance of the Patches [R], were obtained using the X-Rite spectrophotometer. Sanjyot Gindi Masters Thesis Presentation, July 18 th, 2008 Slide 14 of 43

Expt 1: Lamp output spectrum measurement. n The 99% diffuse reflectance target (Labsphere, N. Sutton, NH) was used to reflect the light from the lamp of the scanner ( Setting: ‘Document Feed’ mode) n Spectroradiometer PR-705 (Photo Research Inc. )-was used to measure the spectrum of the lamp by focusing the aperture (1/2 degree) on the white diffusion standard. The spectrum is obtained in the range of 380 nm to 780 nm. n The outputs of the PR-705: q Spectral Radiance (W/sr/sq. m) q X, Y (Luminance in cd/sq. m), Z ; L*, a*, b*, L, u, v, and chromaticity x, y. Sanjyot Gindi Masters Thesis Presentation, July 18 th, 2008 Slide 15 of 43

Schematic of the experimental assembly for lamp response measurement White diffusion standard Lig fro ht re m the flect dif ed fus e clamp Direction of scan head motion r Spectroradiometer (PR-705) Scan-head Lamp Samsung Scanner Sanjyot Gindi Masters Thesis Presentation, July 18 th, 2008 Slide 16 of 43

Lamp Spectrum - Measured by Spectroradiometer. Wavelength in nm Lamp spectrum matrix = [ L ] , 31 x 31 diagonal matrix Sanjyot Gindi Masters Thesis Presentation, July 18 th, 2008 Slide 17 of 43

Expt. 2: Sensor Response. 1. The Monochromator SP-150 was used to input light with wavelengths in the range 400 -700 nm in steps of 10 nm. 2. During the measurement, the lamp of the scanner was off. 3. Scanner setting: Gamma ‘on’ and shading ‘off’, 150 dpi. 4. The Assembly used for measurement was as shown in the following figure: Sanjyot Gindi Masters Thesis Presentation, July 18 th, 2008 Slide 18 of 43

Cross-section view of scan head and measurement assembly Fiber optic cable of the Monochromator Clamp to hold the optic cable glass Scanner head To CCD sensor Lamp (turned ‘off’ during measurement) Incident Light Mirror Lens Sanjyot Gindi Masters Thesis Presentation, July 18 th, 2008 Slide 19 of 43

5. The sample output images obtained are as shown below: 1. 700 nm 2. 540 nm 3. 460 nm 6. For the sensor response, the R, G, B values were averaged over the length of the page and 3 -4 pixels across the breadth of the page. Sanjyot Gindi Masters Thesis Presentation, July 18 th, 2008 Slide 20 of 43

Wavelength in nm Responses of the three channels were taken separately with different intensity settings of Monochromator- 25 for red, 50 for green, 60 for blue. Sanjyot Gindi Masters Thesis Presentation, July 18 th, 2008 Slide 21 of 43

Intensity calibration of Monochromator: n The White diffusion standard was used to reflect the monochromator output and the spectroradiometer measured this reflectance. n The monochromator output measured by the spectroradiometer for all wavelengths from 400 -700 nm is plotted. n Based on the maximum output (found to be obtained at 690 nm) the scaling factor is obtained for all wavelengths n The scaling factor at l= (Intensity at 690 nm)/( intensity at l) n The output of each wavelength is multiplied with this scale factor to obtain the calibrated response. Sanjyot Gindi Masters Thesis Presentation, July 18 th, 2008 Slide 22 of 43

Output variation is from 4. 3 e-5 for 400 nm to 1. 2 e-3 for 700 nm (in units of Radiance)~ approximately 27 times Sanjyot Gindi Masters Thesis Presentation, July 18 th, 2008 Slide 23 of 43

Sensor Response curves n The RGB values are linearized n For each channel, the response < 1% of the maximum value is considered = 0. n The monochromator calibration is applied to each channel Sanjyot Gindi Masters Thesis Presentation, July 18 th, 2008 Slide 24 of 43

Expt 3: Measurement of Reflectance spectra of Kodak. Q 60 color patches. n Using X-Rite DTP 70 Spectrophotometer, 2 degree observer, D 65 illuminant. n Using Spectroradiometer setup, 1/2 degree observer to measure the Kodak Q 60 target illuminated by daylight setting using a Macbeth Spectra. Light II viewing booth Sanjyot Gindi Masters Thesis Presentation, July 18 th, 2008 Slide 25 of 43

Kodak Q-60 target Sanjyot Gindi Masters Thesis Presentation, July 18 th, 2008 Slide 26 of 43

Measurement of Kodak Q 60 patches by X-Rite and Spectroradiometer (Sample # 1) Wavelength in nm X-Rite, 2 deg Observer, D 65 illuminant Sanjyot Gindi Spectroradiometer, ½ deg observer, Illuminant: ‘Daylight’ Masters Thesis Presentation, July 18 th, 2008 Slide 27 of 43

Sample # 2 Wavelength in nm X-Rite Sanjyot Gindi Wavelength in nm Spectroradiometer Masters Thesis Presentation, July 18 th, 2008 Slide 28 of 43

Sample # 3 Wavelength in nm X-Rite Sanjyot Gindi Wavelength in nm Spectroradiometer Masters Thesis Presentation, July 18 th, 2008 Slide 29 of 43

Spectral Model of a Scanner : Lamp [L] R channel G channel B channel Target kodak Q 60 reflectance [R] Sanjyot Gindi Sensor [F] Masters Thesis Presentation, July 18 th, 2008 Slide 30 of 43

Based on the above model: Let [S] = [R] * [L] * [F], then [X] = [M] * [S] L = 31 x 31 diagonal lamp spectrum matrix. R = 240 x 31 matrix- reflectance spectrum of patches on Q 60 Target. F = 31 x 3 sensor sensitivity function matrix. S = 3 x 240 matrix of scanner output data obtained from the model. M = 3 x 3 calibration matrix X = 3 x 240 matrix of CIE XYZ values of the same patches n Matrix M was obtained by simple least squares approximation as [M] = ([S] * [S]T )-1* [S] * [X]T Sanjyot Gindi Masters Thesis Presentation, July 18 th, 2008 Slide 31 of 43

Results: n Regression based method: q Mean Delta E = 3. 82 q Max Delta E = 18. 20 q histogram of number of patches with Delta E values. Sanjyot Gindi Masters Thesis Presentation, July 18 th, 2008 Slide 32 of 43

n Model Based Method q Mean delta E = 4. 205 q Max delta E =19. 332 q Histogram: 115 patches with delta E < 3 Sanjyot Gindi Masters Thesis Presentation, July 18 th, 2008 Slide 33 of 43

Color Gamuts: n The regression method was used to characterize HP, Epson along with the Samsung scanner. n Transformation matrices used to plot 3 D gamuts in L*a*b* spaces. This is based on n Plots of L* Slices of the 3 D gamut n Chromaticity diagram n Metrics for comparison: q Gamut Volume q Quantization error for each channel Sanjyot Gindi inputting uniformly spaced samples from the RGB color cube (0 – 255)x(0 – 255) that comprises the output space of the scanner to the inverse model for the scanner. Masters Thesis Presentation, July 18 th, 2008 Slide 34 of 43

3 D Gamut plots L* b* a* Samsung Scanner Sanjyot Gindi Masters Thesis Presentation, July 18 th, 2008 HP scanner Slide 35 of 43

3 D gamut plot Epson Scanner Sanjyot Gindi Masters Thesis Presentation, July 18 th, 2008 Slide 36 of 43

L* Slices in Gamut Samsung HP Horizontal axis: b* values Epson Vertical axis: a* values L* =20 Sanjyot Gindi L* = 30 Masters Thesis Presentation, July 18 th, 2008 Slide 37 of 43

L*=40, L*=50 Samsung HP Horizontal axis: b* values Epson Vertical axis: a* values Sanjyot Gindi Masters Thesis Presentation, July 18 th, 2008 Slide 38 of 43

L*=60, L*=70 Samsung HP Horizontal axis: b* values Epson Vertical axis: a* values Sanjyot Gindi Masters Thesis Presentation, July 18 th, 2008 Slide 39 of 43

L* = 80, L*= 90 Horizontal axis: b* values Vertical axis: a* values Sanjyot Gindi Masters Thesis Presentation, July 18 th, 2008 Samsung HP Epson Slide 40 of 43

Chromaticity diagram 0. 9 y chromaticity Samsung How can sensor chromaticities lie outside the spectral locus? HP 0. 5 Epson 0. 1 0. 0 Adobe RGB 0. 1 Sanjyot Gindi x chromaticity 0. 5 Masters Thesis Presentation, July 18 th, 2008 0. 9 Slide 41 of 43

Chromaticity diagram 0. 9 y chromaticity Samsung HP 0. 5 Epson 0. 1 0. 0 How can sensor chromaticities lie outside the spectral locus? – Scanner is projecting onto a different 3 -D subspace than the human visual subspace. It doesn’t see color the same way as a human being. Adobe RGB 0. 1 Sanjyot Gindi x chromaticity 0. 5 Masters Thesis Presentation, July 18 th, 2008 0. 9 Slide 42 of 43

Gamut Volume n Gamut volume metric from paper by Braun and Spaulding [3]. n Divide the RGB color Lattice into tetrahedrons—Tetrahedral Tesselation n Compute the L*a*b* values of the vertices n Calculate the volume of each tetrahedron in L*a*b* space in cubic DE units given by: Volume = 1/6 * n P 1(L*) – P 0(L*) P 1(a*) – P 0(a*) P 1(b*) – P 0(b*) P 2(L*) – P 0(L*) P 2(a*) – P 0(a*) P 2(b*) – P 0(b*) P 3(L*) – P 0(L*) P 3(a*) – P 0(a*) P 3(b*) – P 0(b*) Where P 1, P 2, P 3 and P 0 represent the vertices of a tetrahedron and ‘| |’ denotes the determinant. Sanjyot Gindi Masters Thesis Presentation, July 18 th, 2008 Slide 43 of 43

Gamut Volumes (in cubic DE units): n Number of tetrahedrons used: 17, 576 n The gamut volumes calculated are as follows: q Samsung: 2, 012, 700 units q HP: 819, 700 units. q Epson Expression: 780, 370 q s. RGB color space: 811, 180 q Adobe RGB color space: 1, 186, 315 Sanjyot Gindi Masters Thesis Presentation, July 18 th, 2008 Slide 44 of 43

Quantization Error [3] n For the set of L*a*b* values of patches on the Kodak. Q 60 target, the number of points within the gamut of each scanner was determined. n For each of these in-gamut points, the R value was incremented by one and the corresponding L*a*b* value was determined. n The average delta E error between the original value and the incremented value is the quantization error for R channel n Similarly repeated for G and B channels. n Thus quantization errors for each channels is determined in delta E units. Sanjyot Gindi Masters Thesis Presentation, July 18 th, 2008 Slide 45 of 43

Quantization Error in delta E units. Name Delta E R channel Mean Max G channel Mean Max B channel Mean Max Samsung 2. 83 12. 81 4. 35 17. 33 3. 15 12. 06 HP 1. 64 9. 66 2. 58 12. 28 1. 96 9. 21 Epson 2. 24 11. 02 3. 30 14. 41 1. 72 7. 65 Sanjyot Gindi Masters Thesis Presentation, July 18 th, 2008 Slide 46 of 43

Summary of the discussion… n The model based method was used to characterize Samsung SCX 6320 scanner n The regression based method was used to characterize the Samsung, HP and Epson scanners n 3 D gamuts for all 3 scanners were plotted n Slices of the 3 D gamut in L* were plotted n Gamut volume and Quantization error calculation for all 3 scanners. Sanjyot Gindi Masters Thesis Presentation, July 18 th, 2008 Slide 47 of 43

References: 1. “Digital Color Imaging Handbook”, Gaurav Sharma. 2. “Color Science Concepts and Methods Quantitative Data and Formulae”, by Wyszecki and Stiles 3. ”Method for Evaluating the Color Gamut and Quantization Characteristics of Output-Referred Extended-Gamut Color Encodings”, Gustav Braun and Kevin Spaulding, Tenth Color Imaging Conference: Color Science and Engineering Systems, ISBN / ISSN: 0 -89208 -241 -0 4. ECE 638, ”Principles of digital color Imaging systems”, notes by Prof. J. Allebach, Purdue University. 5. ECE 637, “Image Processing”, notes by Prof. C. Bouman, Purdue University. 6. "A Review of Linear Color Descriptor Spaces and Their Applications, " M. Wolski 7. "Imaging Colorimetry Using a Digital Camera, " W. Wu et al 8. Non-Contact Imaging Colorimeter for Human Tooth Color Assessment Using A Digital Camera, " D. Ng et al Sanjyot Gindi Masters Thesis Presentation, July 18 th, 2008 Slide 48 of 43

Additional slides…

Red: transformed values Black: actual values Sanjyot Gindi Masters Thesis Presentation, July 18 th, 2008 Slide 50 of 43

Red: transformed values Black: actual values Sanjyot Gindi Masters Thesis Presentation, July 18 th, 2008 Slide 51 of 43

Variation in Spectral Radiance over a time period of 400 seconds Wavelength in nm Sanjyot Gindi Masters Thesis Presentation, July 18 th, 2008 Slide 52 of 43