OP, please read this page and hopefully you will understand why it's not possible to losslessly compress any data. Some data will actually get bigger: http://en.wikipedia.org/wiki/Pigeonhole_principle
Now I have a _very_ simple example for you. Lets say we want to "compress" 2 bits of data. All the possible datasets are:
00
01
10
11
Now please compress those 4 datasets so that they will all get smaller. Impossible. Unless you use something like:
00 -> 0
01 -> 1
10 -> 10
11 -> 11
But then my next question will be:
Now please compress
0
1
Impossible again, since you've already used 0 and 1 to "compress" 00 and 01. So these 2 "datasets" will have to get bigger by definition, or you will have a collision.
The reason why compression works, is that it encodes sequences and often reoccurring patterns. But as we just saw, there will be datasets (random data) that will actually increase in size if a lossless compression algorithm is used on it. And it just so happens that movies and pictures are very similar to random data. Unless, of course, you use raw pictures/movies, but these files are huge. And believe me when I say you won't beat current algorithms which have been developed over the years by a lot of skillful people.
The thing I think you're missing, is this: You are talking about reference points. This dictionary to which these reference points point, also needs to be stored somewhere. And even if you use referencing, you will still run into the Pigeon Hole Principle. 2 bits of data have 4 possible combinations, so in order to be able to reference to those 4 possible combinations you need 4 different reference points.....Which require 2 bits of data.
Hope this helps.
Now I have a _very_ simple example for you. Lets say we want to "compress" 2 bits of data. All the possible datasets are:
00
01
10
11
Now please compress those 4 datasets so that they will all get smaller. Impossible. Unless you use something like:
00 -> 0
01 -> 1
10 -> 10
11 -> 11
But then my next question will be:
Now please compress
0
1
Impossible again, since you've already used 0 and 1 to "compress" 00 and 01. So these 2 "datasets" will have to get bigger by definition, or you will have a collision.
The reason why compression works, is that it encodes sequences and often reoccurring patterns. But as we just saw, there will be datasets (random data) that will actually increase in size if a lossless compression algorithm is used on it. And it just so happens that movies and pictures are very similar to random data. Unless, of course, you use raw pictures/movies, but these files are huge. And believe me when I say you won't beat current algorithms which have been developed over the years by a lot of skillful people.
The thing I think you're missing, is this: You are talking about reference points. This dictionary to which these reference points point, also needs to be stored somewhere. And even if you use referencing, you will still run into the Pigeon Hole Principle. 2 bits of data have 4 possible combinations, so in order to be able to reference to those 4 possible combinations you need 4 different reference points.....Which require 2 bits of data.
Hope this helps.