For open source applications I would recommend scikit image:

http://scikit-image.org/

And imageJ or Fiji (imageJ w/ extra packages standard}

http://imagej.nih.gov/ij/

If the letters that you are are known, then maybe template matching.

Otherwise active contours or active appearance models

Please look into the following, this may help you

www.isca.in/rjrs/archive/v2/i6/1.ISCA-RJRS-2012-372.pdf

Shahzad, Amir, Muhammad Sharif, Mudassar Raza, and Khalid Hussain. "Enhanced watershed image processing segmentation." Journal of Information & Communication Technology 2, no. 1 (2008): 01-09.

See our publications on research gate

Detection of Discontinuities

¾ Point Detection

¾ Line Detection

¾ Edge Detection

3. Edge Linking and Boundary Detection

¾ Local Processing

¾ Global Processing via Hough

Transform

¾ Global Processing via GraphTheoretic Techniques

4. Thresholding

¾ Foundation

¾ The Role of Illumination

¾ Basic Global Thresholding

¾ Basic Adaptive Threshloding

¾ Optimal Global and Adaptive

Thresholding

¾ Use of Boundary Characteristics

for Histogram Improvement and

Local Thresholding

¾ Thresholds Based on Several

Variables

5. Region-Based Segmentation

¾ Basic Formulation

¾ Region Growing

¾ Region Splitting and Merging

6. Segmentation by Morphological

Watershed

More recent methods focus on hybrid / combination of algorithm or optimization of the existing edge/ region/ contours based methods

Image Segmentation is very simple by using the FAO software for example; the GLCN MadCat, and eCognition Developer.