use morphological operators 

Trying to improve this binary image is less desired than getting the segmentation better. Adding or removing a pixel here or there is just cosmetics: the binary image does not provide a physics justification for doing so.

@Lambert

I have applied median filtering before segmentation, median filtering removes noise and preserves edge points ... I think this is the reason of the small  elongated parts !

Dear Nisreen,

I recomend you to use ImageJ or Fiji. Firstly use some plugins to segmentation blood vessel image. After you can useColour Deconvolution (http://www.mecourse.com/landinig/software/cdeconv/cdeconv.html) or Color Segmentation (http://bigwww.epfl.ch/sage/soft/colorsegmentation/) in ImageJ [put the plugins in ImageJ plugin folder] or Trainable Weka Segmentation in Fiji (built in Fiji, http://fiji.sc/Trainable_Weka_Segmentation_-_How_to_compare_classifiers). After you can transform the results in to binary image with Auto Treshold (http://fiji.sc/Auto_Threshold)or with Make binary.

If you have some questions please ask me.

Best wishes,

Ion C. Andronache.

You can apply an image enhancement technique or edge enhancement method before the segmentation

I suggest you use a gaussian filter with fspecial.

http://stackoverflow.com/questions/15854228/matlab-image-sharpening-gaussian-high-pass-filter-using-1-gaussian-low-pass can give you some hints to start.

Techniques presented in the following paper could be of interest for enhancing and smoothing purposes:

R. Su, C. Sun, C. Zhang, and T. D. Pham, "A New Method for Linear Feature and Junction Enhancement in 2D Images based on Morphological Operation, Oriented Anisotropic Gaussian Function and Hessian Information", Pattern Recognition, 47(10):3193-3208, October 2014.  http://dx.doi.org/10.1016/j.patcog.2014.04.024