**5.1 Moving zones detection**

To accelerate the ME process, we have focused on the image zones where there are movements so that we will conduct the motion estimation only in them. Many techniques have been developed to detect the moving zones in an image. The simplest method is to subtract the background by comparing every image pixels displacement to a prefixed threshold and assuming that it belongs to the foreground if it is superior to this threshold and it is declared as a background's pixel otherwise (Spagnolo, 2006). Hence, the foreground is considered as moving zones. This method is not very efficient since it depends essentially on the prefixed threshold. For this, recently, more sophisticated methods have been built to overcome this limit. Criminisi and al (Criminisi, 2006) have developed a bilayer segmentation method based on the calculation of a complex energy function.

In our system, we have used the background subtraction technique develop by Zivkovic and van der Heijden (Zivkovic, 2006) which models every image pixel's colour values distribution with a mixture of Gaussians (GMM). The mean and the covariance of each component in the mixture are updated for each new video frame (image) to reflect the change of the pixel values. In the case when the new pixel value is far enough the mixture, the pixel is considered as a foreground. This method has shown its rapidity and its good segmentation results in a big variety of videos (as shown on Figure.5).

computation requirement is highly reduced and the compression ratio is increasing, but also

The BMA is an efficient method for motion estimation which encourages us to use it in our multiresolution based method. Unfortunately, despite their encouraging proprieties and their promising results, the BMA and DWT suffer from some problems. For this, a several improvement techniques have been implemented to surmount these problems and make

Despite that it outperforms the conventional motion estimation methods, our proposed DWT based method still having some problems. As we have mentioned before, the DWT representation suffers from the problem of aliasing and the fact that it is a shift variant transformation. Moreover, the block based motion estimation causes the blocking effect which gives a discontinuity in the block boundaries of the predicted image. That is what

These techniques make the motion estimation process more precise and more rapid by detecting the moving zones and limiting the estimation operation to it, adding a sub-pixel precision to the motion vector computing, applying the motion estimation to a shifting variants of the original image aiming to make the estimation a shift invariant operation, overlapping the frame blocks to correct the motion vector by their neighbouring vectors and finally, refining the prediction by changing the block size and re-predicting the blocks which are falsely predicted. In this section we will describe these techniques as well as the causes

To accelerate the ME process, we have focused on the image zones where there are movements so that we will conduct the motion estimation only in them. Many techniques have been developed to detect the moving zones in an image. The simplest method is to subtract the background by comparing every image pixels displacement to a prefixed threshold and assuming that it belongs to the foreground if it is superior to this threshold and it is declared as a background's pixel otherwise (Spagnolo, 2006). Hence, the foreground is considered as moving zones. This method is not very efficient since it depends essentially on the prefixed threshold. For this, recently, more sophisticated methods have been built to overcome this limit. Criminisi and al (Criminisi, 2006) have developed a bilayer segmentation method based on the calculation of a complex energy

In our system, we have used the background subtraction technique develop by Zivkovic and van der Heijden (Zivkovic, 2006) which models every image pixel's colour values distribution with a mixture of Gaussians (GMM). The mean and the covariance of each component in the mixture are updated for each new video frame (image) to reflect the change of the pixel values. In the case when the new pixel value is far enough the mixture, the pixel is considered as a foreground. This method has shown its rapidity and its good

segmentation results in a big variety of videos (as shown on Figure.5).

drives us to develop some additional techniques to overcome these problems.

our method maintains a good prediction quality.

our method more robust giving best results.

**5. Additional improvement techniques** 

that conduct us to implement them.

**5.1 Moving zones detection** 

function.

This temporal segmentation based moving zones detection has allowed us to estimate the motion only on limited zones. Thereby, this technique will reduce the computational time of the ME process and gives a more precise estimation with the assumption that the motion vectors of the blocks which are out of the detected zones will have a null value. This gain is increased if the movement is concentrated in very limited zones.
