Skip to main content

Full text of "DTIC ADP010553: Methods for Deriving Optimum Colours for Camouflage Patterns"

See other formats


Defense Technical Information Center 
Compilation Part Notice 


TITLE: Methods for Deriving Optimum Colours for 
Camouflage Patterns 

DISTRIBUTION: Approved for public release, distribution unlimited 

This paper is part of the following report: 

TITLE: Search and Target Acquisition 

To order the complete compilation report, use: ADA388367 

The component part is provided here to allow users access to individually authored sections 
of proceedings, annals, symposia, ect. However, the component should be considered within 
the context of the overall compilation report and not as a stand-alone technical report. 

The following component part numbers comprise the compilation report: 

ADP010531 thru ADP010556 



Methods for Deriving Optimum Colours for Camouflage Patterns 

K.D. Mitchell, C.R. Staples 

Science and Technology Division. 

Defence Clothing and Textiles Agency 
Flagstaff Rd 
C02 7SS 
United Kingdom 



The majority of camouflage patterns have been designed 
subjectively with only the colour aspect conforming to certain 
constraints such as average colour and luminance. Given the 
power of modern computing it should be possible to design 
scenario specific camouflage from calibrated colour imagery. 
The Defence Clothing and Textiles Agency is at present 
working on such a system. This capability will allow us to 
design and test patterns in a digital environment before field 
trials are carried out. This system will allow us to design 
patterns for specific scenarios such as coniferous treelines, 
deciduous treelines, summer, winter etc. It should also lead to 
highly effective patterns, as early validation can be carried out 
using a target detection model followed by photosimulation 
using a digital implantation technique. Once validated in the 
digital environment, a field trial using live observers can be 
carried out. 

In the design of a pattern, there are two major factors to take 
into account: the multi-level structure of a background and the 
many colours present. A method of designing scenario 
specific patterns needs to reduce the many hundreds of colours 
to a workable number of colour centres, usually between three 
and six. There is also the need to assess the structure present 
and produce a structure for the pattern, which should be multi- 
level to allow the pattern to be effective at various ranges. 

In this paper, we will review the results obtained from the 
initial study on reduction of the number of colours and colour 
centre choice. 

Keywords: Colour choice, patterning, optimisation routine 


Traditionally methods of camouflage design for materials have 
been mainly subjective with the only constraints being the 
colours used and the average luminance of the overall pattern. 

The methods of traditional design involve the designer 
viewing a background and using their skill and judgement to 
devise a pattern which will be effective. This pattern must 
then be trialled to assess its effectiveness and may also 
undergo further validation techniques such as photosimulation. 
Any validation routines have by necessity to take place over 
various scenarios and compare several camouflage schemes. 
The personnel and time needed for such validation makes the 
costs very high. The pattern designed often has to be 
applicable to several theatres of operation i.e. temperate zones, 
jungle environments, arctic and desert and must be a good 
average to account for the diversity within each background. 


The generation of patterns from digital images using 
computers gives the capability to design scenario specific 
patterns, relatively quickly and cheaply. There are three 
discrete parts to the design of a new camouflage pattern using 
a digital methodology. 

1 . A method of texture analysis and generation 

2. A method of optimising the choice of colours from those 
found in a background so the pattern is most effective 
either against a specific background or over a wide range 
of scenarios. 

3. A target detection model which will allow us to measure 
the relative effectiveness of camouflage schemes 

Parts 1 and 3 can be carried out using either colour or 
monotone images but, for an effective visual camouflage part 
2 is a highly important factor. 

A methodology which allows us to carry out textural analysis 
could also be used to design a pattern which has first and 
second orders statistics that resemble those of the background. 
The in-service U.K. pattern has an average colour which 
resembles an average colour of a set number of treelines. The 
design only incorporates the first order statistics of a series of 
backgrounds. Second order statistics are used to describe the 
textural elements of the particular region being analysed. This 
ability to design the first and second order statistics of the 
pattern and the use of target detection models allows us to 
predict the relative effectiveness of several patterns in a digital 
environment. This reduces the initial costs, as we do not have 
to go through such a large-scale trial and do not have the 
expense of making life-size uniforms or designs for vehicles. 
The use of colours in a pattern can determine its effectiveness. 
A method of optimising the limited number of colours used is 
highly desirable. 


The colours in a camouflage pattern or scheme are ideally 
used to allow the target to blend in to the background. This is 
done on two levels. Firstly, the colours used are those found 
in the background. For rural camouflage, these are browns, 
greens and black resembling those found in a natural scene. 
Secondly, the colours form a pattern which, it is hoped match 
that of the background and reduces any visual cue given by the 
outline. It may be said that the colours and textured pattern 
used in a camouflage scheme are equally important from a 
detection point of view. A good pattern’s effectiveness will be 
reduced by bad colour choice and good colours will be 
ineffective if the patterning is poor. It should be noted that no 
matter how good the colours or pattern at very long ranges 

Paper presented at the RTO SCI Workshop on “Search and Target Acquisition”, held in Utrecht, 
The Netherlands, 21-23 June 1999, and published in RTO MP-45. 


both arc inconsequential e.g. at long ranges where the 
background appears monochromatic and atmospheric effects 
dominate 1 . At closer ranges, the better the patterning choice 
and colours used, the shorter the detection range. However, 
for vehicles in particular, there are ranges where the vehicle 
just cannot be disguised. 


The human eye can sec all the colours in a specific 
background but has a problem if asked to reduce these colours 
to a given number for a best fit. The human eye tends to blend 
the colours it sees where as a digital image taken at close 
range will only average over a very small area, depending on 
how the digitisation is done. In addition, humans tend to have 
a bad visual memory for exact colours, whereas a calibrated 
digital image will contain exact data for a specific scene at a 
given moment in time. As a result, choosing the colours to be 
used to optimise the effectiveness of a pattern is a task more 
suited to digital calculation than to human judgement. 


For our colour rcduction/optimisation routine, we decided on a 
methodology which concentrated on a particular Region Of 
Interest (ROI). The values which describe the colour of each 
pixel, in a 3D colour space, arc run through a mathematical 
routine which finds the best fit colour centres for the colour 
population of the ROI. 

Before using the routine, decisions have to be made as to how 
its various capabilities arc going to be utilised. It is necessary 
to decide how many colour centres are to be used, and whether 
a number of those centres are to be predetermined or all are to 
be optimised. 

The first step in the use of the actual program is the section of 
a ROI from the image. If we use the whole image, which 
might be up to 4000x4000 pixels, the length of time needed to 
run the routine may extend into a number of weeks. A good 
size for a ROI is up to 200x200 pixels (although this will have 
to be run overnight if a large number of colour centres is to be 
used). The size of the ROI is up to the individual user, but it 
should contain a good cross-section of the colours found in the 
background as well as some of the textural elements. Figure I 
shows a region which is 100x100 pixels in size and contains 
good information on the type of background we want to be 
camouflaged against. 

Fig 1: ROI taken from original image 100x100 pixels 

As stated, once the region of interest has been chosen, the 
routine then converts the RGB values for each pixel to the Lab 
values. The conversion to Lab colour space allows us to 
describe the colours in a colourspace which resembles how the 
colours are actually perceived by humans 2 . This conversion is 
described in more detail in Houlbrook 3 . Once the conversion 
has taken place the values arc plotted as in Figure 2. This plot 
allows the operator to view' the most populated volumes and 
so place the initial colour centres near these population 

centres. This allows the routine to run quicker and ensures 
that in the later stages all of the colour centres are 

Fig 2: Plot of the Lab values of the population of the 
pixels found in Fig 1 

The next step of the routine is an iterative step, which ceases 
when the best colour centres are found. During the initial step 
of the iteration process, each of the pixel points interrogates 
the initial colour centres and assign themselves to their 
respective nearest colour centre. Figure 3 shows a 
simplification of this initial assignment process. 


Fig 3: Showing an example of Pixel co-ordinates (•) in 
colourspace being assigned to colour centres 1 (A) and 
2 (A) 

Once the co-ordinates of the pixels have all been assigned to 
the most appropriate colour centre an average is taken of the 
population to find the centre point. This centre point is 
assigned as the new colour centre (Sec Figure 4) 

Fig 4: After step one the colour co-ordinates are 

assigned to their respective colour centres (• and •) and 
the colour centres relocated to the average of the 
population ( colour centre I 'A and colour centre 2'A ). 


This stage is repeated with all of the pixel points interrogating 
each of the new colour centres and the averaging process 
repeated. The iteration process ceases when the difference 
between the newest colour centre and the previous colour 
centre is within a predefined limit. These final colour centres 
are then written to file for retrieval later. The final step of the 
routine is to change the pixel co-ordinate values to those of the 
colour centre to which it is assigned. So the whole population 
of that colour centre has the same Lab values. The routine 
will then show a visual representation of the original image 
with the new Lab values. This allows the operator to do a 
visual comparison as a check. Although the human visual 
system is not good at remembering exact colours when asked 
to compare two images, it can judge quite effectively if the 
colours chosen appear to be correct or not. Figure 5 gives an 
example of this phenomenon, in that although there are 
approximately 9000 fewer colours used than in Figure 1, the 
overall impression is not greatly diminished in the reduction to 
10 colours. 

Figure. 5: Visual representation of results of the routine 
(i) shows optimisation to 10 colours (ii) shows 
optimisation to 4 colours. 


As an initial test of the routine, it was decided to use simple 
images constructed in Adobe PhotoShop. The rationale 
behind using images synthesised in PhotoShop is that images 
will be initially viewed using this software. Using PhotoShop 
we can create images with known Lab and RGB values. This 
allows us to compare the Lab values in PhotoShop to the 
values obtained from the routine when digital RGB is the only 
input. This gives us a good insight to how good the routine is 
at calculating Lab’s from RGB. 

The image in Figure 6 consists of 66 pixels and was created so 
we could carry out calculations both manually and using the 
routine to check that the initial conversion was correct. 

(i) (ii) 

Fig 6: Image constructed to check conversion from 
digital rgb to Lab (i) and image obtained at end of routine 




















































Table 1: Comparison of the Lab’s quoted in PhotoShop 
and those obtained as results from the routine 

As can be seen from Table 1 the results obtained from the 
routine are similar to those quoted in PhotoShop. The 
differences can in part be accounted for by rounding errors. 


In addition to the synthesised imagery, the colour reduction 
routine has been successfully applied to real imagery. The 
results when applied to real imagery are shown in Table 1. 
The original image is shown in Fig 1. (The printing process 
has degraded these images to a certain degree.) 

It may be asserted however that the colour-reduced imagery of 
real scenes is, by eye, a good match for the original imagery. 
Where we allow 10 colour centres, the scenes are barely 
discernible from the originals. This fidelity naturally falls 
with the number of allowed colour centres. 

This match to the background is not, however, the purpose or 
test of the routine. That will come with the application of 
chosen colours in the creation of new camouflage schemes. 
These will be tested for their ability to blend into the 
background - by detection modelling, photo-simulation and 
ultimately, test in the real world 


As mentioned there are several variations which can be carried 
out using the colours reduction routine. Firstly we can carry 
out the optimisation routine which will allow us to reduce the 
number of colours in an image to a given number. These 
colours will also be optimised to best describe the colours in 
the original image. Secondly, when inputting the starting 
colour centres we can choose to lock these centres so that the 
colour reduction takes place to these colour centres. This is 
particularly useful if the colours have been pre-determined and 
it is these colours you want to use. You can now see how 
those colours compare to those in the background. We also 
obtain information on the proportion of each colour in the 


We have described in this paper a routine, which can be used 
to derive optimum colours for a camouflage pattern using 
calibrated digital imagery. It has been recognised that the 
optimisation of colours for a pattern is desirable. Optimised 
colours used in a pattern can reduce the ranges at which 
targets become visible in specific scenarios. 


Colour as has been described is an intrinsic part of a 
camouflage pattern. In the rush to devise a digitally based 
method for pattern design, the use of the best colours has been 
largely overlooked. This work addresses this oversight and 
represents an important step in the development of a more 
complete digital pattern design tool. A routine such as this 
allows the optimisation of colours for a scenario without the 
need for extensive field trails and so cuts the time needed for 
the design of an effective pattern for that scenario. 


1 Phillips, P.L., “Colour Tolerances for Texture 
Investigation”, Final BAe report on MoD Contract No. 
A82a/2782. 1983. 

2 McDonald, R., “Colour Physics for Industry'”, Society of 
Dyers and Colourists, 1997. 

3 Houlbrook, A.W., “The Development of an Image 
manipulation Facility for the Assessment of CCD”, This 
Proceedings, 1999.