160 likes | 271 Views
From the Warehouse to the Camera to the Computer: Finding Accurate Representations of Fabric Colors. Roxana Trofin and Stephanie Claussen Psych 221Final Project March 20, 2007. Project goal. To determine a method for correctly displaying the color of clothing on a sRGB monitor
E N D
From the Warehouse to the Camera to the Computer: Finding Accurate Representations of Fabric Colors Roxana Trofin and Stephanie Claussen Psych 221Final Project March 20, 2007
Project goal • To determine a method for correctly displaying the color of clothing on a sRGB monitor • To display a fabric’s “true color” using its measured spectral power density • To use the Macbeth Color Checker to understand the processing of an image that is performed by the camera that captures it • To correct for any misrepresentations of color introduced by the camera and obtain the “true color”
Motivation Customers bought $9.6 billion worth of clothing online in the United States in 20061 • But online purchases represent only 5% of overall clothing sales Customers return 30% of the clothes they buy online Comments from Jo-Ann Fabric and Craft website (joann.com) "Colors not accurate on monitor" …If I were to order on-line again, I would probably call and confirm the actual description of the color prior to ordering. The color I thought I was ordering was "nowhere" near the color I received. Just be careful. "Alpine Fleece" Their color charts do not do justice to the colors of the fleece. The ivory is nice and neutral, the alpine green is a lovely rich deep green. 1. New York Times, “Log In Your Measurements, and the Clothes May Fit”, 03/12/07
From clothing to display sRGB values are put through the monitor gamma and displayed Camera measures linear R,G, and B values of fabric Color balancing B. Wandell., Psych 221 course notes, “Image capture” J. Farrell, Psych 221 course notes, (3/07/07)
Sources of error • Monitor assumptions • sRGB “standard” color space • Gamma function • Information loss due to only 3 channels on camera • Illumination estimation • Color balancing algorithms • Gray World • White Balancing Tungsten Daylight Fluorescent
Investigation #1: How do you most accurately represent a color? Multiply by the SPD of D65 light to see what the fabric would look like under a standard illuminant Remove the effect of the illuminant to determine the fabric’s absolute SPD Using a spectrophotometer, measure the SPD of the fabric and the light used to illuminate it Convert linear sRGB values to digital frame buffer values assuming a gamma of 2.2 Calculate the XYZ values and convert to sRGB values Apply normalization and display the color on your sRGB monitor!
Results of Investigation #1 “TRUE” COLOR Orange fleece “TRUE” COLOR Blue linen
Investigation #2: How do you “fix” an image captured by a camera? Introducing… the Macbeth Color Checker! A photo of the Macbeth Color Checker, when taken under the same conditions as the fabrics, gives you all the information you need to correct for any color misrepresentations http://www.as-found.net/exhibitions/gretagmacbeth/
Process for using the Macbeth Color Checker to correct fabric images (1) Determine the gamma of the camera by using the bottom 6 linear squares on the MCC to calculate the linear RGB values of the MCC Take a picture of the MCC under the same lighting conditions as your images Obtain a matrix of the RGB values of the MCC Find the correction matrix that maps the RGB values from the camera MCC to those of the standard Obtain the standard linear RGB values of the MCC
Process for using the Macbeth Color Checker to correct fabric images (2) Tungsten TRUE COLOR TRUE COLOR Fluorescent TRUE COLOR TRUE COLOR Daylight TRUE COLOR TRUE COLOR
Results of Color Correction Images made better- green linen Comparison to joann.com Original image Corrected image Tungsten illuminant setting TRUE COLOR Fluorescent illuminant setting Images made worse- yellow satin Comparison to joann.com Original image Corrected image Tungsten illuminant setting TRUE COLOR Fluorescent illuminant setting
Analysis of results Error = average of R, G, and B errors with respect to the “true” RGB values found in Investigation #1 Successful color correction was achieved for the fleeces and linens Satin and velvet were difficult fabrics to correct
More analysis of results Similar trends were found for fabrics with the camera set to fluorescent and daylight illumination settings
Sources of error • Data gathering • Images of Macbeth Color Checker were taken at a different angle than that of the fabrics • Could give an incorrect 3x3 correction matrix • Spectrophotometer and camera (and the lighting used) were at different angles when the sets of data were taken for the two investigations • Fabric differences • Different fabrics reflect differently • Irregular lighting could affect fabrics differently • Averaged the errors in the R,G, and B values
Conclusion • Knowing the reflectance spectrum allows you to most accurately display a color on a monitor • A picture of a Macbeth Color Checker taken under the same condition as the picture of interest allows the image to also be accurately processed and displayed by the online retailer • Clothing websites could also post this Macbeth Color Checker on their site to enable their customers to see an accurate portrayal of the item they’re purchasing
Acknowledgements Thank you to Professor Wandell and Jon Winawer A special thank you to Joyce Farrell, for all of her help defining a project, starting us down the right path, and helping keep us along that path!