General Considerations:

In my opinion, the most important (and difficult) aspect of DIC is to impart an appropriate pattern on the sample being tested. From my experiences, patterning has the most profound impact on the quality of results, so most of the effort in obtaining good data should be centered around imparting a good pattern first.

Also, all patterns, regardless of length scale, should be spatially random. In this sense, the use of the word "pattern" is sort of a misnomer, but it is the phrase most commonly used in literature so I use it here. The randomness ensures the DIC tracking algorithms work properly.

Large Macroscopic Length Scales (2cm FOV and above):

This length scale is the easiest to pattern. Samples in this length scale are generally patterned with white and black matte (non-reflective) spray paint by hand. I've also seen some samples which are patterned through printing with a silk screen or transfer paper, but by far the most common method is to simply spray paint the surface. An example of a simple spray paint pattern is shown in the next section; an example of the silk screen printing process can be found in the paper below:

G. Stoilov et al, "A Comparative Study of Random Patterns for Digital Image Correlation." Journal of Theoretical and Applied Mechanics (2012).

Small Macroscopic Length Scales (2-5mm FOV):

I've done some experimentation in this length scale and one process that seemed to work was to use a filter in conjunction with spray paint. The filter acted to remove large paint particles while allowing smaller ones to pass through and deposit on the surface. A diagram of a filter box is shown below:

The idea is to place a small sample on the permeable grill, place a filter (I used a standard house air filter) in the filter seat, and then attach a vacuum to the vacuum port. This will draw the spray paint through the filter, and then deposit small paint particles on the sample surface. The difference in the results obtained from directly spray painting and using the filter box are shown below:

One negative aspect of this process is that the particles had poor adhesion. The paint particles most likely dried before they reached the surface of the sample. One way I dealt with this was to apply a thin layer of silicone glue to the surface of the sample, before applying the spray paint, which caused the particles to adhere. The problem with this approach is that the particles are not directly adhered to the surface, meaning they will not deform with the sample, although they will still move as the sample deforms. 

Another idea I implemented to improve adhesion was to use acrylic spray paint, and then set the sample above a bath of acetone briefly (~10 seconds) before removing it. Acrylic dissolves readily in acetone (which is highly volatile). The idea is illustrated below:

Once again, there are trade-offs to this method. The act of placing the sample over an acetone bath improves adhesion, but causes the particles to spread, which is illustrated on the right. This reduces the efficacy of the pattern at this length scale because the particles become larger.

Microscopic Length Scales (500-1000µm FOV):

This length scale is typically patterned by using a high quality airbrush with a very fine tip which atomizes the paint. One example is the Iwata custom micron B which I've used in my lab (a publication is also provided below for an additional example) and is shown in the picture above, along with an air compressor. This provides highly atomized speckles for smaller length scales. An additional use of the air brush is that any liquid that has low viscosity can be sprayed through it (including colloidal suspensions, high temperature paints, etc).

In my personal experience, I actually found that the best results for atomizing paint finely is to collect regular spray paint (in something like an eppendorf tube) and then spray it through the air gun. Spray paint has very good adhesive properties as well as very low viscosity. This resulted in the finest atomization in the paints I tested. Furthermore, I also found that a pressure of around 30 PSI range resulted in the best atomization in the pressure ranges I tested. A summary of the results is shown below:

T.A. Berfield et al, "Micro- and Nanoscale Deformation Measurement of Surface and Internal Planes via Digital Image Correlation." Experimental Mechanics (2007).

Nano Length Scales (nm FOV):

In this length scale, the most popular method to pattern samples appears to be with the use of nanoparticle mixtures or colloidal dispersions. Usually, a nanoparticle solution is applied to a sample surface through simple drop casting. In other cases, I've also seen structures fully submerged in a colloidal solution before being extracted.

A.D. Kammers et al, "Small-scale Patterning Methods for Digital Image Correlation Under Scanning Electron Microscopy." Measurement Science and Technology (2011).