Rajendra,
This has to do with how link state protocols work. In order for a link protocol to be able to run the Dijkstra algorithm, they must all have identical databases to run the algorithm against. If filtering were allowed within an area, then by definition, some routers would have a different database than others, and when those routers ran the shortest path tree calculations, they would arrive at different results.
Notice that both the ASBR and ABR, by definition, are at the area boundary. They are allowed to do filtering or summarization only on routes that exist outside of the area where they are presenting the filtering or summarization. This ensures that all routers within the area where the ASBR or ABR has done filtering will all be getting the same information (from the ASBR or ABR).
To put it another way, imagine an ABR like this:
(AREA 0)—ABR—(AREA 1)
The ABR is NOT allowed to summarize or filter an area and re-introduce that back into the same area. So it is okay for the ABR to filter Area 0 and present that to Area 1, or summarize area 1 and present the summary to area 0.