Psy 221 Project Report

Conclusion


For all the images we have considered, the block matching algorithm does not produce better quality images than the median or bilinear algorithm. Subjectively, we cannot see any improvement in the image quality. More objectively, the signal to noise ratios (SNR) are not consistently higher than those of the median and bilinear algorithm. Computationally, the block matching algorithm is more complex than all the other algorithms we considered. It consumes more computer time for the estimate computations.

Our best advice for digital camera algorithm design would be to stick with the median or bilinear algorithm if we were to choose only from the algorithms we have considered for this project. The hardware requirement for the block matching algorithm will almost certainly impose high costs to manufacturing of the chips for digital cameras.

We should point out that SNR (or MSE) does not satisfactorily measure the image reproduction quality, because a difference of 10% or even 20% in SNR may not at all translate into any difference in perceivable image quality. And, by some construction, we can also conceivably create images that have similar SNR but of drastically different qualities. Such measure (SNR or MSE) of the image reproduction quality does not take into account of the peculiar properties of the human visual system. There is still no known image reproduction quality measure that incorporates the characteristics of the human visual system.

For future projects, we would like to consider some edge detection and pattern recognition algorithms. With such algorithm incorporated, we should be able to reduce a lot of the artifacts on sharp edges (as manifested in Example two). Such adaptive algorithms likely will greatly increase the computational complexity.