Study of filter effects in wavelet image compression

Abstract

Extended abstractIn this paper, we study the filters' effects in Image compression by wavelet transform. The principle of wavelet transform is to decompose hierchically the input image into a series of successively lower resolution reference images and detail images which contain the information needed to be reconstructed back to the next higher resolution level.The histogram of image sub-bands provides us with information on the distribution of the coefficient values in this sub-image. The sub-band images resulting from wavelet transform are not equal significance. Some sub-bands contain more information than others. The total number of available bits describing an image is however inevitably limited. Therefore, it is desirable to allocate more bits to those sub-bands images which can be coded accurately than others. The objective of a such bit allocation method is to optimize the overall coder performance and minimize the quantization error. In determining which wavelet filter is to be used for image compression, some of the properties considered are vanishing moments. The phase non-linearity of the filter can cause severe degradation in the subjective quality of an image. It is related to the symmetry of the filter coefficients. The wavelet transform is implemend using a linear-phase. Biorthogonal filter with four levels of decomposition.For this study, we use a scalar quantization with uniform threshold quantizers. The quantization method is PCM (Impulse Coded Modulation) for the coefficients in all high-pass sub-bands. The coefficients of low-pass sub-bands are DPCM (Differential PCM) quantized per region.Suitable criteria are needed to evaluate rigorously the performance of e compression scheme. In case of images, the search for simple and suitable criteria is hindrcd by the fact that the results obtained by statistical performance criteria may not agree with subjective evaluation of the humain eye. Since the objective is data compression, the compression ratio at optimal distorsion is obviously an important performance measure. The reconstructed imae quality can be evaluated by mean-squared error (MSE) and peak signal to noise ratio (PSNR).We can subdivide the image in one or mode sub-images ; if we increase the level of decomposition, the compression quality ("PSNR) will improve.From this study, we can conclude that the PSNR is very similar when the level of decomposition is greater than 4. The biorthogonal filters provide us good results in image compression.Finally, the filters' effects, like those of the decomposition levels, on the compression of the images by wavelets, are very significants. The best choice of the filter and decomposition levels can ensure us a best quality of compression.

Image Sequence Measures for Automatic Target Tracking

“…This definition makes sure that the IFCDTS equals 0 when the target size between adjacent frames is constant, and equals 1 as the target size change range between adjacent frames is equal to the target size itself. Being similar with IFCDT, we force IFCDTS values between 0 and 1 and modify Equation (19) to…”

“…Many of the global gray-level metrics in the literature seem to be designed with that in mind. The standard deviation of an image and the entropy of its histogram are two of the simplest measures of an image [17][18][19].…”

“…3. An image sequence metric focuses on measuring the change degree of target size is developed in this section, and it is calculated as follows: (19) where N is the number of images in the given image sequence, and l i and w i are the length and width of the smallest rectangle containing a target in the i-th image frame, respectively. This definition makes sure that the IFCDTS equals 0 when the target size between adjacent frames is constant, and equals 1 as the target size change range between adjacent frames is equal to the target size itself.…”

“…Since the characteristics of a single image and the characteristics of an image sequence are different, it is not surprising that the metrics for single images are not best suited for measuring image sequences. The existing image sequence measure methods, such as peak signal-to-noise ratio (PSNR) [16,17], mean square errors (MSE) [18][19][20], mean structural similarity (MSSM) [21], have been proposed in the fields of image coding and image communication. These metrics, different from that in the field of automatic target tracking, focus on measuring information losses in video compression, therefore they have not been applied to evaluate the performance of automatic target tracking algorithms in image sequences.…”