Deblock filter

[angelzhang]

hi,everyone
In the JM8.6 software,the deblock filter calculates value of boundary strenth for 16 samples in four 4*4 blocks, which takes large timming cost. and now, i want to cut down timing cost by reducing the calcultion loop. In my opinion, boundary strenth values of four samples in one block are same. then we can just calculate one sample other than four times. am i right?
please give your suggestions!

best regards
angel

[hworldjj]

Not sure, I think you can check this idea on JM8.6. But I am afraid there will be a little degradation of subjective quality, especially for low resolution frames.

[angelzhang]

thanks for your reply.
Then what do you think a good way to solve the question of much timming cost?

I just run the modified software ,but can't find much change about SNR or picture display performance.

then, i am confused how to know the effect about my modification.

angel

[Manao]

angelzhang : your change isn't specs compliant ( since you can get a result different from the JM's one, if I understood you correctly ). Of course, in a quality vs. speed tradeoff, your change may be totally valid, but bear in mind you must respect the specs.

And finally, in the JM, last time i heard of, deblocking was clearly not the slow part of the computation.

[hworldjj]

PSNR means nothing for deblock filter. Maybe you should test more sequences, especially low bit rate and low resolution, where artifacts will be observed easily. In addition, bitstreams encoded by your codec will not be complimant to H.264 standard, so decoder will accumulate error.

[angelzhang]

acutally, the process of deblock filter in Encoder is idetical to that in Decoder. then the PSNR in the decoder must be same as that in encoder. but i'm confused that how to improve the performance of part of deblock filter. or we can do nothing about the part.

best regards
angel