The formation of desiccation cracks on soil surface is a common natural phenomenon as it is subjected to drought climate. The presence of cracks can significantly affect the hydro-mechanical behaviour of soil, and results in various geotechnical problems. With increasing frequency of severe drought climate, better understanding of soil desiccation cracking behavior is becoming an increasingly significant issue. In this investigation, laboratorial desiccation tests were performed on soils by simulating long term drought climate. A camera was mounted above the specimen to monitor the initiation and propagation behaviour of desiccation cracks. Image processing technique was employed to quantify and characterize the obtained crack patterns. It is observed that the desiccation cracking generally takes place in three stages: main-cracks firstly start on soil surface and form main-clods; subsequently, main-clods are split into several sub-clods by sub-cracks; after the size of all clods is stable, cracking terminates and the final crack pattern is formed. It is shown that the image processing technique is efficient to accurately determine the geometrical parameters of crack patterns during drying, including crack width, length, clods area, etc. The introduced density function is effective in describing the distribution characteristics of the geometrical parameters of crack patterns.