| 摘要 |
<p>The present invention relates to a method for compressing data without loss, in which training data, comprising of a set of symbols of an alphabet, are used to construct a tree machine of a certain number of levels comprising a root node at the highest level, nodes at lower levels, and branches between nodes in two adjacent levels. An identifier is defined for each branch representing a symbol or a set of consecutive symbols of the training data. A context is defined for each node representing a string of the symbols of the branches from the node to the root node. Further, at least one count is defined for each node representing a probability distribution of symbols following the context of the node. In the method a difference value is defined. Also for a node other than the root node a first code length (L(s)) and a second code length (I(s)) are defined. The first code length (L(s)) for a leaf node is defined by using the probability distribution of the node itself. The first code length (L(s)) for an internal node is defined by using the probability distribution of the node itself and/or the probability distribution of at least one child node of the internal node. The second code length (I(s)) is defined by using the probability distribution of the parent node. If said second code length (I(s)) is smaller or equal than the sum of said first code length (L(s)) and said difference value, the node under examination, and all the child nodes of the node under examination are removed from the tree machine.</p> |
