The industry's recognized authority for design, engineering and application of equipment and services in the global ocean community
Issue link: http://sea-technology.epubxp.com/i/99539
Every COTS digital camera has its own RGB color space because each sensor has unique spectral sensitivities. This difference is usually substantial between different makes of cameras and between different models of the same make. Even two copies of the same camera make and model produced at the same factory are likely to have slightly different spectral sensitivity curves. Transforming raw images recorded by COTS digital cameras to a deviceindependent color space is usually done through a three-by-three transformation matrix that relates the camera color space to the human color space. While cameras have built-in software that perform this transformation, they do so without the knowledge of the colors in the scene, the ambient light conditions or the spe(Top Left) A collage of features representative of three dif- cifc sensor in the camera. raphy is not appropriate for two reasons. First, these targets contain color patches selected to represent the range of colors humans see on land on a daily basis and, thus, are more diverse and saturated than the colors found in most underwater habitats. Second, the transformation process for land-based photos frequently uses the spectra of standard illuminants, which do not ferent underwater habitats. Photos were taken at a variety of depths (all shallower than 10 meters) at different times Transformation Matrix represent underwater ambient of day. (Photo Credits: Derya Akkaynak, Justine J. Allen Colors will only be as accurate as light felds well, such as the and Elron Yellin) the color transformation matrix. The International Commission on CIE 1931 XYZ color space is comIllumination (CIE) D-series of (Above) Pictured left, the Ocean Optics USB2000 singlemonly used as a device-independent illuminants for approximating channel spectrometer coupled with a handheld computer in a custom-made underwater housing. A major disad- space into which camera RGB valthe natural daylight spectra ues are mapped. The color-matching (D65 is used as the baseline in vantage of this system is that downwelling and sidewelling profles cannot be collected simultaneously. Pictured functions (CMF) associated with this many applications). model approximate the spectral senOne solution is to create right, a new system, also by Ocean Optics, under development contains eight spectrometers with probes at fxed sitivities of an average human with habitat-specifc color charts, locations. Three probes on top measure polarization, in normal color vision. If known, the or habitat charts. This requires addition to fve probes that measure downwelling and CMFs of an animal can be used to collecting refectance data sidewelling light at various angles. (Photo Credits: Derya simulate the way colors would be from a variety of substrates Akkaynak) perceived by that visual system. This found in a given underwater should, however, be done with caution, as COTS digital habitat with a spectrometer, as well as representative light cameras are designed to produce images that look pleasing profles. The ideal way to collect spectral data in an unto the human eye and have three spectral sensitivity curves derwater scene is by using a hyperspectral imager, which that peak in the regions of the visible part of the electromagrecords a spectrum for every pixel in an image within a netic spectrum that humans perceive as red, green and blue. certain range of the electromagnetic spectrum. However, All COTS digital cameras are tuned specifcally to the hyperspectral imagers are expensive and therefore not availthree color receptors in the cones of human eyes. However, able to many research laboratories. some animals are tuned to different colors (or even have four In their absence, spectrometers can be used. This way, color receptors for increased color sensitivity), and some are images captured by COTS digital cameras can be transable to see in the ultraviolet or infrared parts of the elecformed from the camera space to a device-independent tromagnetic spectrum in which a COTS camera records no space (such as standard RGB) using colors and light spectra information. that are representative of that particular habitat. Each habitat The transformation matrix M from the camera RGB space has different optical properties of water and varying colors to the XYZ space is derived as follows: RGB x M = XYZ. amid the plants, animals and abiotic entities of each subFor land photography, M is frequently computed using strate. Therefore, unique habitat charts should be built for a Macbeth ColorChecker. The refectance spectra of each each dive site. patch and their XYZ tri-stimulus values corresponding to different illuminants are published. For underwater applicaRaw Images tions, the XYZ values are obtained from the refectance and When the intensity of pixels in an image constitutes scilight spectra recorded for a particular habitat feature as folentifc data, the analysis should start with raw images rather lows: than the post-processed fles (e.g., JPG or TIFF). Post-processed fles will not only be irreversibly altered in features Equation pdf like white-balance, color and contrast but may also be compressed in a lossy fashion. Such changes that happen withIn this equation, i = 1, 2, 3..., and λmin and λmax are the limits out user control cannot be quantifed and will compromise of the visible range, R(λ) is the refectance spectrum of the patch data quality. www.sea-technology.com DECEMBER 2012 / st 11