JPH06253281A - Picture signal orthogonal transformation method - Google Patents

Abstract

PURPOSE:To concentrate transformation coefficients onto a base of a low frequency with less calculation processing when signals at an optionally shaped area in a picture are orthogonally transformed. CONSTITUTION:A picture analysis section 2 analyzes input picture data from a terminal 1 and outputs data 21 comprising lateral picture elements (a) and longitudinal picture elements (b) of a square area V surrounding a processing object picture segment W, W shape data 23, and W picture data 24. A base arithmetic operation section 3 calculates aXb sets of bases with respect to the area V and outputs a base picture in which the bases are arranged sequentially from a lower frequency in both longitudinal and lateral directions. A base section 4 applies zigzag scanning to the base picture from a low frequency to select n-set of base objects. A projection transformation section 5 projects the base objects onto the segment W and a base orthogonal section 6 makes orthogonal processing. DCT is applied to the picture data 24 by using the orthogonal bases for bases of the segment W and a transformation coefficient is outputted to a terminal 7.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image signal orthogonal transformation method for orthogonally transforming a signal of an arbitrary shape region in an image in an image coding method for efficiently transmitting or accumulating an image signal.

[0002]

2. Description of the Related Art A method of orthogonally transforming and coding a signal in an arbitrarily shaped region in an image is, for example, "Coding of Arbitrary".
Shaped Image Segments Based on a Generalized
Orthogonal Transform "(M. Gilge, et al, Sign
al Processing: ImageCommunication Vol.1, 19
90). Here, inter-frame prediction is performed on a moving image signal, orthogonal transformation is performed on an arbitrary shape region W whose prediction error value is greater than or equal to a predetermined threshold value, and the transform coefficient is quantized and encoded. The method of conversion is shown. Further, in this document, as a method for obtaining an orthogonal transformation basis for an arbitrary shape region W consisting of n pixels,
A × b orthogonal transformation bases in a rectangular region V, which is composed of horizontal a pixels and vertical b pixels surrounding the arbitrary shape region W, are obtained, and the orthogonal transformation bases are projected onto the arbitrary shape region W to newly obtain a ×
Although it is described that b bases are recalculated, n bases are selected from these bases, and orthogonal transformation bases are calculated by reorthogonalization, there is no specific method for selecting a × b bases. Not mentioned.

On the other hand, a conventional technique for selecting n bases from a × b bases is, for example, "Segment-based Im".
age coding by Shape Orthogonal Transform ”
(Y. Kato, Picture Coding Symposium '91, 1
2.4-1, 1991). This method normalizes a × b bases projected on an arbitrary shape region W, and when the modeled image signal is converted based on these bases, n bases are selected in order from the base having the highest energy concentration. It is to select. When n bases selected by this method are re-orthogonalized, the degree of energy concentration on the orthogonal bases is greatly biased. Therefore, the orthogonal transform coefficient is quantized and coded to efficiently convert the image signal. Can be encoded as

[0004]

According to the above conventional technique, a
In order to select n bases from xb bases, the image signal must first be modeled, and the energy concentration degree for each base must be calculated using the signal.
The calculation has to be performed every time the shape of the region is different, which causes a problem that the amount of calculation becomes very large.

The object of the present invention is to select n bases from a × b bases with a small amount of calculation, and to concentrate the conversion coefficients on the low-frequency bases to the same extent as in the above conventional method. An object of the present invention is to provide a method for orthogonally transforming a signal in an arbitrarily shaped region consisting of n pixels in an image.

[0006]

In order to achieve the above object, the present invention provides an arbitrary shape region W consisting of n pixels in an image.
A × b in a rectangular area V of horizontal a pixels and vertical b pixels surrounding the
Number of bases, the orthogonal transformation bases are projected onto the arbitrary shape region W, a * b bases are newly obtained, n pieces are selected from these bases, and the orthogonal transformation bases are obtained by re-orthogonalization. At this time, a horizontal pixel and a vertical b surrounding the arbitrarily shaped region W
When obtaining a × b orthogonal transformation bases in a rectangular region V made up of pixels, horizontal a × vertical b are arranged in the order of a base representing a low frequency component to a base representing a high frequency component in each of the horizontal and vertical directions. When n bases are arranged and n bases are selected from the bases obtained by reprojecting these bases on the arbitrary shape region W, the horizontal a
× The top n bases are simply scanned in a zigzag order from the base that represents the lowest frequency component in the horizontal and vertical directions to the base that represents the highest frequency component in the b vertical bases. Is to be selected.

[0007]

It is possible to arrange the bases in order from the low frequency component to the high frequency component at the same time as obtaining the orthogonal transform base. In addition, since image signals generally have energy concentrated in low-frequency components, by simply scanning in a zigzag order from the basis representing the lowest frequency component and selecting the upper n basis thereof. Even without using a model of an image signal, it is possible to bias the degree of energy concentration in each base to the same extent as in the above-mentioned conventional method, and efficient encoding can be performed.

[0008]

An embodiment of the present invention will be described in detail below with reference to the drawings.

FIG. 1 shows a case where a signal in an arbitrary shape region in an image is orthogonally transformed and the orthogonal transformation coefficient is quantized / encoded.
It is a block diagram which shows one Example of the conversion base derivation | leading-out part which concerns on this invention. In this embodiment, DCT is used as the orthogonal transform. In the figure, 1 is an input terminal, 2 is an image analysis unit, 3 is a DCT basis calculation unit, 4 is a base selection unit,
Reference numeral 5 is a projective transformation unit, 6 is a base orthogonalization unit, and 7 is an output terminal. Further, 21 is a target segment W that is an arbitrarily shaped region
Data indicating vertical and horizontal lengths of a rectangular area V surrounding the
Is data indicating the number of pixels forming the target segment W, 2
Reference numeral 3 indicates data representing the shape of the target segment W, and reference numeral 24 indicates image data such as a luminance component in the target segment W. Hereinafter, the operation of FIG. 1 will be described step by step.

Image data is sent to the input terminal 1.
In this example, it is assumed that the prediction error image data obtained by taking the inter-frame difference is sent. This image data enters the image analysis unit 2, is divided into several segments, and the area is analyzed for each segment. From this image analysis unit 2, a rectangular area V surrounding the segment W is displayed.
Data 21 indicating the length and width of the segment W, data 22 indicating the number of pixels forming the segment W, and data 23 indicating the shape of the segment W are sent out as information regarding the area. In this example, the segment W as shown in FIG.
Was included. 2 shows the target segment W
A rectangular area V surrounding 11 is a total of 132 pixels in the horizontal direction and 12 in the vertical direction, and the target segment W is a partial area including 67 pixels.

The data 21 is sent to the DCT basis calculation unit 3 and the DCT basis for the rectangular region V is calculated. When obtaining the DCT bases, horizontal a × vertical b bases are arranged in the order of a base representing a low frequency component to a base representing a high frequency component in each of the horizontal and vertical directions. FIG. 3 shows, for the example of FIG.
The DCT bases of a total of 132 pixels are arranged so that the vertical and horizontal directions are arranged in order from the low frequency range, and one square corresponds to each base image.

The base image thus obtained is passed to the base selecting section 4 together with the data 22 indicating the number of pixels (67 pixels in FIG. 2) forming the target segment W. The base selection unit 4 scans the base image of the rectangular region V in a zigzag manner from the base representing the lowest frequency component in the horizontal and vertical directions to the base representing the highest frequency component, and the upper segment W thereof is scanned. Select as many bases as there are pixels to configure. In this example, 67 basis candidates are selected in the zigzag scan order as shown by the arrow in FIG.

The 67 basis candidates thus selected are sent to the projective transformation unit 5. The projective transformation unit 5 refers to the data 23 showing the shape of the target segment W and refers to 67
Project the bases onto the segment W. Since the orthogonality of the projected base is generally broken, the base is further sent to the base orthogonalization unit 6, where the orthogonalization procedure is performed. For orthogonalization, for example, the Schmidt method or Householder
Method.

As described above, 67 finally obtained.
The orthogonal base composed of these elements is used as the base of the target segment W, and DCT is applied to the image data 24 such as the luminance component in the segment W sent from the image analysis unit 2. The conversion coefficient thus obtained is output from the terminal 7.

[0015]

As described above, according to the present invention,
When orthogonally transforming a signal in an arbitrarily shaped region in an image, the transform coefficients can be concentrated on the low-frequency base to the same extent as in the conventional method with less calculation processing, and efficient encoding is possible.

[Brief description of drawings]

FIG. 1 is a block diagram of an embodiment of the present invention.

FIG. 2 is a diagram showing an example of an arbitrarily shaped area W and a rectangular area V surrounding it.

FIG. 3 is a diagram for explaining arranging DCT bases in the vertical and horizontal directions from a low band to a high band and selecting a base by zigzag scanning.

[Explanation of symbols]

 1 Input Terminal 2 Image Analysis Section 3 DCT Basis Operation Section 4 Basis Selection Section 5 Projection Transformation Section 6 Base Orthogonalization Section 7 Output Terminal

Claims (1)

[Claims] 1. A method of orthogonally transforming a signal of an arbitrary shape region in an image, wherein a × b is set for a rectangular region V of horizontal a pixels and vertical b pixels surrounding an arbitrary shape region W consisting of n pixels in the image. Number of orthogonal transformation bases are obtained, the a × b orthogonal transformation bases are projected onto the arbitrary shape region W, a × b bases are newly obtained, and the newly obtained a × b pieces are obtained. From the n pixels by selecting n bases from the bases, re-orthogonalizing the selected n bases to obtain n orthogonal bases, and using the re-orthogonalized orthogonal bases. A method of orthogonally transforming a signal of an arbitrary shape region W, wherein when obtaining a × b orthogonal transform bases for the rectangular region V, a base higher than a base representing a low frequency component in each of a horizontal direction and a vertical direction is obtained. The order of bases representing frequency components is a ×
When n bases are selected from new axb bases re-obtained by arraying b bases and projecting the axb orthogonal transformation bases onto the arbitrary shape region W, Of the b bases, the top n bases are selected by scanning in a zigzag order from the base representing the lowest frequency component in the horizontal and vertical directions to the base representing the highest frequency component. An image signal orthogonal transformation method characterized by: