CN100468442C

CN100468442C - Method for performing multi-object image identification without recording full image

Info

Publication number: CN100468442C
Application number: CNB2006100035097A
Authority: CN
Inventors: 陈美如
Original assignee: Pixart Imaging Inc
Current assignee: Pixart Imaging Inc
Priority date: 2006-02-05
Filing date: 2006-02-05
Publication date: 2009-03-11
Anticipated expiration: 2026-02-05
Also published as: CN101013468A

Abstract

It's an image identification method without a record of the whole image, which can do real-time identification to number of objects. And the objects come from multiple image segments. The method includes the following steps: (A) Capture video image pixel value per column in order; (B) Judge unknown object image segments starting point; (C) Accumulated points with the image information of point-by-section. (D) Judge segment end of unknown objects in the image; (E) Use this image in the region and space-related adjacent segments to decide which object the image segments belongs to (F) Pool the image segments to accumulate information of the object; (G) Go for next section judgment of this section images; (H) After capturing all the images at the pixel, identify the object of the image.

Description

Do not need the recording full picture can carry out the method for multi-object image identification

Technical field

The invention relates to a kind of image recognition (Image recognition) method, be meant a kind of method that can handle, not need the recording full picture can carry out the multi-object image identification in real time especially.

Background technology

Present image processing techniques, object (Objects) for identification any amount in image, often need to cooperate different image edge sharpening arithmetic, and general algorithm is along with the increase of desiring identification objects quantity in this image, it is complicated that the calculating of its algorithm also more becomes, for example must use complicated region growing (Regiongrowing) algorithm, also therefore must in advance the frame buffer (Image buffer) of whole image (full images) in image processing system be stored, after collecting the information of all images, just can carry out complicated recognizer to pick out each object in this image, thus, not only take the memory source of frame buffer during identification, also very expend time in.

Summary of the invention

Therefore, purpose of the present invention promptly can be carried out the method for multi-object image identification at the recording full picture that do not need that provides a kind of and need not to use frame buffer, can save memory source.

Another object of the present invention is promptly providing a kind of and is having extendibility, not limited by the number of objects in the image, and the recording full picture that do not need that all can pick out each object in real time can carry out the method for multi-object image identification.

So, the present invention does not need the recording full picture can carry out the method for multi-object image identification, image taking sensor cooperates buffer, this imageing sensor has a plurality of determinant inducing pixels, and described determinant inducing pixel is to carry out identification by the mode of row induction in real time for a plurality of objects that have in the image, each described object can be made up of and each image section is made up of the inducing pixel of delegation at least the image section of at least one row, and described image identification method comprises following step: the pixel value that (A) captures every row in the image in regular turn; (B) in the described image of acquisition, in the pixel value of every row, judge the image section starting point of unknown object in these row; (C) from the information of the image section of the described unknown object of starting point pointwise accumulative total of the image section of described unknown object; (D) judge the image section terminal point of unknown object described in these row; (E) image section of utilizing unknown object described in these row and the spatial coherence of the image section of adjacent each object of previous column differentiate this be listed as described in the affiliated object of image section of unknown object; (F) object of the information that image section added up of compiling described unknown object under it; (G) when there are other unknown objects in these row, carry out the judgement of the image section of the next unknown object in these row; When (H) having captured once all pixel value of described image, promptly pick out each the described a plurality of object in the described image simultaneously.

Description of drawings

Fig. 1 is that explanation the present invention does not need the recording full picture can carry out the circuit block diagram of the applied image processing system of preferred embodiment of the method for multi-object image identification;

Fig. 2 is that explanation the preferred embodiment is the synoptic diagram that the object of any amount that has for image carries out identification; And

Fig. 3 is each flow chart of steps of explanation the preferred embodiment.

The primary clustering symbol description

1 image, 3 image processing systems

11 circular object, 31 imageing sensors

111～114,121～125 image section, 311 pixels

111a, 112a, 121a, 122a starting point 32 analog-digital converters

111b, 112b, 121b, 122b terminal point 33 graphics processing units

12 triangle object, 34 buffers

101～108 steps

Embodiment

About aforementioned and other technology contents, characteristics and effect of the present invention, in the DETAILED DESCRIPTION OF THE PREFERRED of following conjunction with figs., can clearly present.

As shown in Figure 1, the present invention does not need the recording full picture can carry out the embodiment of the method for multi-object image identification, be applied in the image processing system 3, this image processing system 3 has imageing sensor (Imagesensor) 31, analog-digital converter (A/D Converter) 32, graphics processing unit (Imageprocessor) 33 and buffer (Register) 34, imageing sensor 31 is that CCD or cmos component are made, take the image formation by rays of thing (figure does not show) reflection in order to induction, and be converted to simulating signal; Then, export analog-digital converter 32 to and be converted to digital signal, be responsible for the computing of most of signal by graphics processing unit 33.

Mandatory declaration be that the image processing system 3 of present embodiment can be used for shooting with video-corder the discriminating function that picture waits image-taking device, among other embodiment, maybe can be to carry out discriminating function in the mode of the recognition software that is installed in computing machine; In addition, because the aufbauprinciple of imageing sensor 31, analog-digital converter 32, graphics processing unit 33 and other associated component is a known technology, and main concept of the present invention is the discriminating function that cooperates buffer 34 carries out image with graphics processing unit 33, therefore following will only being described with regard to the part that is relevant to the principle of the invention.

Cooperate shown in Fig. 1,2, the method that the present invention does not need the recording full picture can carry out the multi-object image identification is that the object of any amount that had for the image 1 that imageing sensor 31 is responded to carries out identification, in the present embodiment, in this image 1 to be identified to as if be the step that example explanation is carried out identification with circular object 11 and triangle object 12.

Mandatory declaration be, because imageing sensor 31 has a plurality of determinant inducing pixels (Pixel) 311, and these pixels 311 are to respond to each

object

11,12 by the mode of row, therefore, object resulting parts of images in each row that imageing sensor 31 is sensed is called image section (ImageSegment), circular object 11 for example shown in Figure 2 has the image section 111～114 of four row, and triangle object 12 has the image section 121～125 of five row, by that analogy.

Cooperate shown in Fig. 1～3, details are as follows not need the recording full picture can carry out each step of method of multi-object image identification and action principle the present invention:

At first, capture the pixel value (step 101) of every row in the image 1 in regular turn, that is from left to right read each pixel value in these row, constantly read each pixel value of every row by that analogy from the first row beginning from imageing sensor 31; When reading, the image section starting point of judging unknown object in these row wherein and be stored to buffer 34 (step 102), starting point from this image section begins, the information of this image section of pointwise accumulative total also is stored to buffer 34 (step 103), judges the image section terminal point of this unknown object in these row again and is stored to buffer 34 (step 104); And step 102 judges whether that to 104 the mode of object images information is to detect the pixel value that whether has greater than the systemic presupposition threshold value to occur; Then, utilize this row image section and adjacent previous column respectively the spatial coherence of the image section of this object image section of differentiating these row belong to what object (step 105).

In the present embodiment, for judging as meet following formula 1 that then this unknown object image section of decidable belongs to this object i:

Seg-L ≦ Preline-Obji-R; And

Seg-R ≧ Preline-Obji-L; Formula 1

Wherein, when supposing to read to the y column data in the image 1, then Preline-Obji-R represents that y-1 lists the right-hand terminal point X coordinate figure of the image section of existing respectively this object i; Preline-Obji-L represents that y-1 lists the image section left starting point X coordinate figure of existing respectively this object i; Seg-L represents to read the image section left starting point X coordinate figure that y lists existing unknown object; Seg-R represents to read the right-hand terminal point X of the image section coordinate figure that y lists existing unknown object.

After compiling information that this image section the adds up object (step 106) under it, in like manner, carry out the judgement (step 107) that this is listed as next image section, then captured all pixel values of image after, also finish the multi-object identification (step 108) of this image diverse location simultaneously.

Cooperate shown in Fig. 1,2, suppose in image 1, to begin to read each pixel by row from the 1st row, because coordinate (3,1) locates to occur greater than the pixel value of systemic presupposition threshold value, therefore the coordinate figure of starting point 111a that just writes down object 11 in buffer 34, then pointwise accumulative total image section 111 information and be stored in the buffer 34, up to the terminal point 111b that runs into this image section 111, the coordinate figure that writes down this terminal point 111b again is in buffer 34; Yet owing to then have the information of another image section 121 in first row, the information that also need store the coordinate figure of starting point 121a, terminal point 121b of this image section 121 and pointwise accumulative total thereof again is in buffer 34.

Then, record is listed as the left starting point coordinate value 112a of each unknown object image section of (the 2nd row) appearance again, 122a, and the right-hand terminal point coordinate value 112b of each unknown object image section, 122b, and when reading to the right-hand terminal point of each unknown object image section, whether belong to each object 11 as criterion to distinguish with formula 1 immediately, 12, with right by a left side, order from top to bottom is a principle, by the row record, computing is listed as to last, therefore, when having captured the value of these image 1 all pixels 311, can pick out respectively this object 11 in this image in real time fully, 12.

Conclude above-mentionedly, the method that the present invention does not need the recording full picture can carry out the multi-object image identification has following advantage:

1. image identification method of the present invention is carried out image identification in real time in temporary mode, need not to use frame buffer, therefore can save memory source.

2. image identification method of the present invention is not limited by the number of objects in the image because algorithm is simple, therefore any amount of object all can be come out by identification, not only has extendibility, also can pick out each object in real time.

The above person only is the preferred embodiments of the present invention, can not limit scope of the invention process with this, and promptly all simple equivalent that claim and invention description content are done according to the present invention change and modify, and all belongs in the scope that patent of the present invention contains.

Claims

1. method that does not need the recording full picture can carry out the multi-object image identification, image taking sensor cooperates buffer, this imageing sensor has a plurality of determinant inducing pixels, and described determinant inducing pixel is to carry out identification by the mode of row induction in real time for a plurality of objects that have in the image, each described object is made up of the image section of at least one row and each image section is made up of the inducing pixel of delegation at least, and described image identification method comprises following step:

(A) capture the pixel value of every row in the described image in regular turn;

(B) in the described image of acquisition, in the pixel value of every row, judge the image section starting point of unknown object in these row and be stored to described buffer;

(C) from the information of the image section of the described unknown object of starting point pointwise accumulative total of the image section of described unknown object and be stored to described buffer;

(D) judge this row described in unknown object the image section terminal point and be stored to described buffer;

(E) utilize the spatial coherence of image section with the image section of adjacent each object of previous column of unknown object described in these row, differentiate this be listed as described in the affiliated object of image section of unknown object;

(F) object of the information that image section added up of compiling described unknown object under it;

(G) when there are other unknown objects in these row, carry out the judgement of the image section of next unknown object in these row; And

When (H) having captured once all pixel value of described image, promptly pick out the described a plurality of objects in the described image simultaneously.

2. the method that does not need the recording full picture can carry out the multi-object image identification according to claim 1, wherein, in the step (E), the mode of affiliated object of differentiating the image section of described unknown object is, when judgement meets following formula, then the image section of the described unknown object of decidable belongs to object i:

Seg-L ≦ Preline-Obji-R; And

Seg-R≧Preline-Obji-L；

Wherein, when reading to the y column data in the image, Preline-Obji-R represents that y-1 lists the right-hand terminal point X coordinate figure of the image section of each existing described object i; Preline-Obji-L represents that y-1 lists the left starting point X coordinate figure of the image section of each existing described object i; Seg-L represents to read the left starting point X coordinate figure that y lists the image section of existing described unknown object; Seg-R represents to read the right-hand terminal point X coordinate figure that y lists the image section of existing described unknown object.