ADV601LC Просмотр технического описания (PDF) - Analog Devices

Номер в каталоге

Компоненты Описание

производитель

ADV601LC

Ultralow Cost Video Codec

Analog Devices 

ADV601LC

THE WAVELET KERNEL

This block contains a set of filters and decimators that work on

the image in both horizontal and vertical directions. Figure 6

illustrates the filter tree structure. The filters apply carefully

chosen wavelet basis functions that better correlate to the broad-

band nature of images than the sinusoidal waves used in Dis-

crete Cosine Transform (DCT) compression schemes (JPEG,

MPEG, and H261).

An advantage of wavelet-based compression is that the entire

image can be filtered without being broken into sub-blocks as

required in DCT compression schemes. This full image filtering

eliminates the block artifacts seen in DCT compression and

offers more graceful image degradation at high compression

ratios. The availability of full image sub-band data also makes

image processing, scaling, and a number of other system fea-

tures possible with little or no computational overhead.

The resultant filtered image is made up of components of the

original image as is shown in Figure 3 (a modified Mallat Tree).

Note that Figure 3 shows how a component of video would be

filtered, but in multiple component video luminance and color

components are filtered separately. In Figure 4 and Figure 5 an

actual image and the Mallat Tree (luminance only) equivalent is

shown. It is important to note that while the image has been

filtered or transformed into the frequency domain, no compres-

sion has occurred. With the image in its filtered state, it is now

ready for processing in the second block, the quantizer.

Understanding the structure and function of the wavelet filters

and resultant product is the key to obtaining the highest perfor-

mance from the ADV601LC. Consider the following points:

• The data in all blocks (except N) for all components are high

pass filtered. Therefore, the mean pixel value in those blocks

is typically zero and a histogram of the pixel values in these

blocks will contain a single “hump” (Laplacian distribution).

• The data in most blocks is more likely to contain zeros or

strings of zeros than unfiltered image data.

• The human visual system is less sensitive to higher frequency

blocks than low ones.

• Attenuation of the selected blocks in luminance or color com-

ponents results in control over sharpness, brightness, contrast

and saturation.

• High quality filtered/decimated images can be extracted/created

without computational overhead.

Through leverage of these key points, the ADV601LC not

only compresses video, but offers a host of application features.

Please see the Applying the ADV601LC section for details on

getting the most out of the ADV601LC’s sub-band coding

architecture in different applications.

NL

MK

I

F

J

H

C

G

E

A

D

B

BLOCK A IS HIGH PASS IN X AND DECIMATED BY TWO.

BLOCK B IS HIGH PASS IN X, HIGH PASS IN Y, AND DECIMATED BY EIGHT.

BLOCK C IS HIGH PASS IN X, LOW PASS IN Y, AND DECIMATED BY EIGHT.

BLOCK D IS LOW PASS IN X, HIGH PASS IN Y, AND DECIMATED BY EIGHT.

BLOCK E IS HIGH PASS IN X, HIGH PASS IN Y, AND DECIMATED BY 32.

BLOCK F IS HIGH PASS IN X, LOW PASS IN Y, AND DECIMATED BY 32.

BLOCK G IS LOW PASS IN X, HIGH PASS IN Y, AND DECIMATED BY 32.

BLOCK H IS HIGH PASS IN X, HIGH PASS IN Y, AND DECIMATED BY 128.

BLOCK I IS HIGH PASS IN X, LOW PASS IN Y, AND DECIMATED BY 128.

BLOCK J IS LOW PASS IN X, HIGH PASS IN Y, AND DECIMATED BY 128.

BLOCK K IS HIGH PASS IN X, HIGH PASS IN Y, AND DECIMATED BY 512.

BLOCK L IS HIGH PASS IN X, LOW PASS IN Y, AND DECIMATED BY 512.

BLOCK M IS LOW PASS IN X, HIGH PASS IN Y, AND DECIMATED BY 512.

BLOCK N IS LOW PASS IN X, LOW PASS IN Y, AND DECIMATED BY 512.

Figure 3. Modified Mallat Diagram (Block Letters Correspond to Those in Filter Tree)

–4–

REV. 0

Share Link: