ISET Selecting a Color Conversion Matrix

Transcription

1 ISET Selecting a Color Conversion Matrix Contents How to Calculate a CCM...1 Applying the CCM in the Processor Window...6 This document gives a step-by-step description of using ISET to calculate a color correction matrix (CCM). The method described here is the default used by ISET. It is based on (a) simulating the responses to a Macbeth Color Checker under a specific illuminant, and (b) using the simulated sensor responses to derive a 3x3 linear matrix (a CCM) that can be used for Manual Matrix Entry. During the process, several graphs and values of the color accuracy provided by the CCM are calculated. The methods described in this application note can be used either for a simulated Macbeth Color checker, calculated from the Scene Window, or from a data set loaded from a file say an image collected by a sensor in the lab. In the example illustrated here, we load the sensor data from a file. If you are working on a simulation, simply simulate the response by creating a Macbeth Color Checker in the Scene Window, set the optics and sensor parameters, and compute the simulated response. How to Calculate a CCM We illustrate the calculation using an image of Macbeth Color Checker acquired using a Nikon D70 under a D65 light. To obtain these data, we read the raw image from the Nikon D70 using Nikon Capture. Then, using the ISET script s_nikond70_sensor, we convert the NEF file into a Matlab file containing the raw sensor data from the Nikon. The script creates the data file in the form of an ISET sensor structure. The NEF file is large, so we do not include it in the ISET distribution. 1

2 Once we have built the Matlab file with the ISET-Sensor structure, we use the pull-down command ISET-Sensor File Load sensor (.mat) to bring the data into the ISET-Sensor window. Figure CCM-1. The raw data from a Nikon D70 are loaded into the sensor window. The image shown here is a cropped version of the image of a Macbeth Color Checker obtained under a D65 illuminant. 2

3 Figure CCM-2. The color conversion matrix, L, needed to transform an MCC image into a linear srgb display is calculated using the pull-down menu shown above. Next, we use the pull-down menu under ISET-Sensor Analyze Color Color Conversion Matrix. This initiates an interaction in which the user uses the left mouse button to select the outer corners at the white, black, cyan and brown patches in the color checker. ISET uses the user-selected points to locate the R,G and B data from the sensor structure and forms a matrix, M, with twenty four rows containing the Macbeth patch RGB values. 3

4 Figure CCM-3. The sequence of steps to identify the color patch locations is shown. Then, ISET creates a matrix, D, containing the RGB values of a linear srgb display. Finally ISET computes a 3x3 matrix, L, that solves, in the least squares sense, D = M*L. This matrix is printed at the display prompt L = You can copy the values of this matrix and use them as the Manual Matrix Entry option in the ISET-Processor window (see below). 4

5 At the end of the color calculation process, an ISET graph window will appear that plots the difference between desired and observed RGB values, along with the CIELAB E color metric. Desired (r,g,b) Count Color error Mean de ab Observed (r,g,b) Delta E ab Figure CCM-4. The desired and observed (calculated) RGB values for a Color Checker when using the derived color correction matrix, L. In this case, there is good agreement in the RGB representation (on the left). There is also good agreement in the CIELAB color space, with a mean color difference of less than 3 CIELAB E units. 5

6 Applying the CCM in the Processor Window Apply the 3x3 color correction matrix to the processed (demosaiced) image. Bring up the Processor Window and select Manual Matrix Entry. Click on the Compute Display Image A small window will appear. Cut and paste the optimal CCM coefficients from the Matlab command window into this window. The image that appears in the Processor Window represents the best one can do, given known and optimized CCM coefficients. 6