Abstract: In the application of image segmentation based on fast level set algorithm, there exist difficulties in level set initialization and setting thresholds, so a new algorithm which combining PYRamid model, Random Walk and Level Set (PYR-RW-LS) is proposed. First, the multi-scale analysis technique is introduced into Random Walk (RW) algorithm, and its partition result is taken as the initialized curve of the fast level set algorithm, so the fast level set algorithms initialization problem is solved; Then the evolution of the level set can be seen as the constant pattern classification of the points on the curve. Both Bayesian classification rule and minimal distance classification rule were introduced by this new algorithm to work alternatively, in order to acquire the driving force for curve evolution. And the invalidation conditions for both of the classification rules are set as the iteration stop conditions in this new algorithm, thus solving the difficulties in setting thresholds. Simulating experimental results show that PYR-RW-LS not only runs faster than the fast level set algorithm, which only adopts pattern classification ideas, but also has better capabilities than RW algorithm in terms of anti-noise capabilities; And the advantages of being insensitive to blurry boundaries remains with the RW algorithm. PYR-RW-LS algorithm, therefore, is good in particular, for images with large size and high resolution.