5 Replies Latest reply on Apr 24, 2008 8:01 AM by (Brian_Carlick)


      I read a message from Lou that mentioned "table based" and "matrix based", could some explain what they are, I remember doing matrix algebra at school....



      A sunny Aldeburgh
        • 1. Re: Table/Matrix
          Lou Dina Level 3

          Table based profiles use lookup tables inside the profile to handle conversions from one color space to another. Table based profiles are the norm with printer profiles, many LCD monitor profiles, and most non-linear devices. Depending on the number of patches read during profiling, there could be as few as perhaps 100 points, or many thousand points, so lookup table based profiles can be moderate to large in size. Of course, not all possible colors are covered, even in the largest profiles, so the color management engine (ICM, ACE, etc) have to interpolate between points. Also, table based profiles support Perceptual, relative colorimetric, saturation and absolute colorimetric rendering intents.

          Matrix based profiles use TRCs (tone reproduction curves) to define the color space and handle conversions. Most editing spaces, CRT monitor profiles, and sometimes profiles for very linear printing devices are matrix based. These profiles are smaller in size and can be very smooth since they are curve based, but are only really suitable for devices that are fairly linear. In an editing space, the selection of the red, green and blue primaries determines the size of the overall gamut, which is sort of triangular in shape. Small editing spaces, such as ColorMatch and sRGB have endpoints that aren't generally "way out there", so the "triangle" these endpoints define is rather small, so they may clip a lot of printable colors. Still, even sRGB has points that cannot be printed on most devices. Adobe RGB is a moderately sized editing space and extends the primaries that define the space, so it has a wider overall gamut, but also has more unprintable colors. Even Adobe RGB clips some printable colors, but does a good job for colorful images.

          To create a matrix based editing space that will encompass ALL printable colors, you need something like ProPhoto RGB, which has primary endpoints that are WAY out there. You do not clip any printable colors with ProPhoto, but at a price. You have a huge swath of colors that are not printable in ProPhoto, conversions can result in some real surprises and shifts, and you must work in 16 bit. Some people like ProPhoto but I am not one of them. In my opinion, it comes with way too much baggage, but to each his own. You can get great photos using ProPhoto, but you need to be more careful and also be aware of the pitfalls.

          PhotoGamut RGB is a table based profile, so it is shaped almost exactly like your typical inkjet printer profile, only larger. It covers nearly all "printable" colors, but does not give you points way out in the ozone, so conversions are more accurate, fewer shifts occur, it tends to be smoother, has more closely spaced points for finer editing, can be used with 8 bit files, etc.

          Personally, I use the smallest editing space I can for a given image...sRGB for average images, Adobe RGB for bright images, and sometimes use PhotoGamut when I have an original that really needs the widest color gamut, such as artwork that I am reproducing, bright saturated yellow daffodils on a sunny day, etc. Some people love ProPhoto, so try it and make your decision, but be sure to work in 16 bit if you do.

          Hope that sheds some light on the subject.

          • 2. Re: Table/Matrix

            thanks for your always profound help. Only this

            > Matrix based profiles use TRCs (tone reproduction curves)
            > to define the color space and handle conversions.
            > Most editing spaces, CRT monitor profiles, and sometimes
            > profiles for very linear printing devices are matrix based.

            requires IMO some further explanations.

            RGB working space profiles and monitor profiles are
            mostly matrix based.
            The profile connection space is CIE XYZ. Special
            understanding of XYZ isn't required - it's just the
            ONE absolute physical color space.

            Any RGB space (as above) is related to XYZ by a matrix:
            R = a11*X + a12*Y + a13*Z
            G = a21*X + a22*Y + a23*Z
            B = a31*X + a32*Y + a33*Z

            But this is valid in a linear RGB space. File data
            are gamma-encoded, using values C':
            C'= R',G',B'
            C = R, G, B
            C = C'^Gamma = C'^2.2 for instance

            These functions are established in ICC profiles
            either as mathematical expressions or as tables.
            These tables are 1-1-tables, but nevertheless
            look-up-tables, called TRCs.

            If an RGB ICC profile is LUT based, then it contains
            3-3-LUTs, in addition to the matrix definition and
            the 1-1-LUTs. Tables with three inputs R,G,B and
            three outputs X,Y,Z, and in the reverse direction,
            but IMO, the essential behaviour is established
            in the 1-1-LUTs and in the matrix, whereas deviations
            are defined in the 3-3-LUT (nonlinear corrections).

            Hopefully true, according to ICC specs.

            Best regards --Gernot Hoffmann
            • 3. Re: Table/Matrix
              Lou Dina Level 3

              Thanks for the in-depth, mathematical clarification and expansion. Hope you are well and keeping busy in your new life.

              Regards, Lou
              • 4. Re: Table/Matrix

                my life isn't much new, but one exception:
                I could convince the landlord of my studio
                for Photography, Color Management and Digital
                Printing to clean the chimney, where a bird
                (the RAVEN ?) had established his nest, so that
                I'm NOW no more affected by CO, CO2, SO2 and
                other unpleasant poisonous gasses. Nevermore.

                Best regards --Gernot Hoffmann
                • 5. Re: Table/Matrix
                  Level 1
                  Thanks Lou and Gernot

                  Some light has inded been shed on the subject, my knowledge has increased again!


                  A sunny Aldeburgh