JPH0527152B2 - - Google Patents

Description

Detailed Description of the Invention [Technical Field of the Invention] The present invention uses two indicators, a correlation coefficient and a density difference for each pixel of two images, to calculate the positional deviation between two images with high density. The present invention relates to a highly practical precision image positioning device that can perform image correction.

[Technical Background of the Invention] In fields such as satellite image processing or medical image processing, when attempting to synthesize a new image from multiple images captured and input with a time difference, it is necessary to Positioning is required. If the characteristics of this positional shift are known, can be considered uniform for the entire image, and can be expressed by a linear coordinate transformation formula such as parallel translation, rotation, enlargement, or reduction, it can be corrected using relatively simple signal processing. It is possible.

[Problems with the background art] However, in the case of the above-mentioned satellite images or medical images, the cause of positional deviation is often a combination of several factors, and the amount of positional deviation changes locally. , it is difficult to align the entire image with a single linear coordinate transformation.

For this reason, conventionally, the image to be aligned is divided into several small areas, and each of these areas to be repaired is moved while calculating the correlation coefficient with the image to be aligned, and this value is maximized. The correct position was the part where the

However, with this method, when the positional shift is minute or the contrast difference between images is small, the amount of change in the correlation coefficient becomes very small and the sensitivity is poor, and when noise is superimposed on the image, this amount of change may be hidden. Because of problems such as convergence sometimes converging on the wrong position, highly accurate alignment was difficult and impractical.

[Object of the invention] The present invention has been made in consideration of overcoming such problems, and its purpose is to:
2. Correlation coefficient and pixel-by-pixel density difference between two images
An object of the present invention is to provide a precision image positioning device that uses two indices to correct positional deviation between two images with high precision.

[Summary of the invention] FIG. 1 shows a diagram of the principle of the present invention.

In the present invention, two images are divided into a plurality of local regions where the amount of positional deviation is considered to be uniform, P ₁₁ , P ₁₂ ,
After dividing P ₂₁ , ..., Q ₁₁ , Q ₁₂ , Q ₂₁ , ..., first calculate the correlation coefficient in the neighboring region for each local region, and then set the correlation coefficient in the image plane at the position where this value is maximum. After performing coarse position correction by repeating movement, when the maximum value of the change in the correlation coefficient in the neighboring area becomes equal to or less than a predetermined threshold, the pixels of the two images in the neighboring area are Precise position correction is executed by repeatedly moving in the direction in which the evaluation function of each density difference is maximized (minimum). Whether the moving direction is the maximum value direction or the minimum value direction is determined depending on how the evaluation function is created.

Initially, the positional deviation is large and the contrast difference cannot be ignored, so the above coarse positional correction is performed using the correlation coefficient, which is a normalized index, and after the positions are almost aligned, the contrast is considered to be almost the same. By executing the above-mentioned precise position correction using the above-mentioned evaluation function as an index, which also takes into account the density difference for each pixel of the two images, the positional deviation is minute and the contrast difference is smaller than when only the correlation coefficient is used. Even in the case of alignment of images with small or slightly different values, it can be expected to prevent convergence to an erroneous position, and this has a great practical effect.

[Embodiment of the Invention] An embodiment of the present invention will be described below with reference to the drawings.

Figure 2 shows the overall configuration.

2 input from the imaging system 1 to the image memories 2 and 3
When aligning two images, the local area dividing circuit 4 first divides each into a plurality of local areas, and then the position correction target area specifying circuit 5 selects a pair of local areas to be aligned. Store in memories 6 and 7.

First, in the coarse position correction section 100, a correlation coefficient is calculated by a correlation coefficient calculating circuit 9 for both images in the local area memory 7 that have been cut out by the local area memory 6 and the neighboring area cutting circuit 8, and a correlation coefficient is calculated. It is stored in the memory 10. Here, the density of each pixel of one local area A to be aligned is a _ij (i
= 1 to m, j = 1 to n), the average value is μ _A ,
The density of each pixel in the other local area B is expressed as b _ij (i=1~
m, j=1 to n), and its average value is μ _B , then the correlation coefficient C _AB of these local area pairs is expressed by the following formula.

After performing this operation for all the neighboring regions specified by the neighboring region extraction circuit 8, the maximum value detection circuit 11 obtains the largest one among these correlation coefficient groups, and the movement position determining circuit 12 calculates this correlation value. The position within the image is moved in the direction that maximizes the relationship coefficient. At the same time, the maximum change amount calculation circuit 13 compares the maximum change amount of the correlation coefficient in the neighborhood region with a predetermined change amount threshold value 14 using a comparator 15, and the maximum change amount becomes equal to or less than the threshold value ( In other words, there is almost no change in the correlation coefficient)
The above coarse position correction is repeated until

If the value falls below the threshold, the coarse/fine switching circuit 16 transfers the information to the fine position correction section 200.

In this embodiment, an absolute value error is used as the evaluation function. In this case, the direction of movement is the direction of the minimum value.

In the precise position correction section 200, the absolute value error between the images in the local area memories 6 and 7 is calculated by the absolute value error calculation circuit 17, and the absolute value error memory 18
Store in.

After performing this operation for all the neighboring regions specified by the neighboring region extraction circuit 8, the minimum value detection circuit 19 determines the smallest absolute value error among these absolute value errors, and the movement position determining circuit 20 calculates this absolute value error. The position within the image is moved in the direction that minimizes .

At the same time, a comparator 22 compares the minimum value of the absolute value error in the above-mentioned neighborhood area with a predetermined absolute value error threshold 21, and the precision position correction completion determination circuit 2
If the minimum value becomes equal to or less than the threshold value in step 3, it is determined that the position has been matched, and the precise position correction is terminated, and the position correction target area designation circuit 5 designates a pair of local areas to be a new position correction target. and repeat the above operation again.

[Effects of the Invention] As described above, first, the coarse position correction section performs position correction using the correlation coefficient as an index, and when the change in the correlation coefficient has almost disappeared, the position correction for each pixel of the two images is performed. By performing precise position correction using the density difference evaluation function as an index, it is possible to achieve simple and highly accurate position alignment between two images, which has great practical effects.

Note that the present invention is not limited to the embodiments described above. For example, in the above embodiment, the absolute value error was used as the precision position correction evaluation function, but more generally, by adopting an evaluation function that combines the density difference for each pixel of two images and the correlation coefficient, it is possible to Highly accurate position correction can be achieved.

In addition, in the coarse position correction section, instead of using the correlation coefficient as an index, each feature point in the image is extracted,
By using a feature watching method to align these positions, even faster alignment can be expected.

In short, the present invention can be implemented with various modifications without departing from the spirit thereof.

[Brief explanation of the drawing]

FIG. 1 is a principle diagram of the present invention, and FIG. 2 is a block diagram of an embodiment of the present invention. 1... Imaging system, 2, 3... Image memory, 6, 7
... Local area memory, 100 ... Coarse position correction section,
200...Precise position correction section.

Claims (1)

[Scope of Claims] 1. An image memory storing two images; local region dividing means for dividing each of the two images into a plurality of local regions; and a plurality of local regions of each of the two images. a position correction target area specifying means for specifying one local area to be aligned from the area; and a position correction target area specifying means for storing one local area specified by the position correction target area specifying means, and storing the other local area including its neighboring area. a local area memory for storing the other local area; a neighboring area cutting means for sequentially moving and cutting the other local area of the local area memory within the neighboring area; and a local area memory for sequentially cutting out the one local area and the neighboring area cutting means. Correlation coefficient calculation means for calculating a correlation coefficient with the other local area, and positioning the other local area and said one local area corresponding to the maximum correlation coefficient calculated by the correlation coefficient calculation means. means for aligning, means for terminating the alignment when the maximum amount of change in the correlation coefficients calculated by the correlation coefficient calculation means is less than or equal to a predetermined threshold; and means for terminating the alignment when the maximum amount of change is greater than the threshold moving the other local region corresponding to the maximum correlation coefficient within the neighboring region in the case;
a coarse position correction means having a means for repeating calculation and alignment of respective correlation coefficients, and another local area aligned by the coarse position correction means and successively cut out by the neighboring area cut-out means and one of the local areas. Absolute value error calculation means for calculating the absolute value error of density for each pixel with respect to the local region, the other local region corresponding to the minimum absolute value error calculated by the absolute value error calculation means, and said one local region. means for terminating the alignment when the minimum absolute value error is less than or equal to a predetermined threshold; and means for terminating the alignment when the minimum absolute value error is greater than the threshold; An image characterized by comprising: precision position correction means having means for repeating movement cutting out of the other local region corresponding to the error within the neighboring region, calculation of each absolute value error, and alignment; Precision alignment device.