In this paper

a series of techniques are developed to address this problem. First

a new concept

adaptive gradient slice(AGS)

is defined for the purpose of transforming a 3D video sequence to a 2D representation. Then image projection and morphological operations are adopted to locate shot transitions in the AGS domain. For the detected transitions

soft templates and corresponding template-matching algorithms are designed to recognize their fine structures. Simulation results show that

with the manually constructed templates

the proposed algorithm can recognize most of the transitions in the experiment very precisely in real time. Only for two transitions quite similar to each other

the proposed algorithm will encounter some difficulties to distinguish them. The reason is that the mapping from 3D sequence to 2D image loses some spatial information. However. In conclusion

this paper provides an effective way to deal with the problem of fine structure recognition of video shot transitions.