JP6849735B2 - Image correction method - Google Patents

Description

The present invention relates to a method of modifying an image that can be observed through an image guide (optical fiber bundle) in which optical fibers are bundled.

In a fiber scope used as an endoscope, an objective lens is placed at the tip of an image guide that bundles optical fibers, and illumination light is emitted from the side of the objective lens via a light guide fiber placed along the image guide. , The image illuminated by this illumination light is captured by the objective lens, imaged at the tip of the image guide behind the objective lens, the imaged light is sent to the eyepiece via the optical fiber, and directly through the eyepiece. It is a structure that can be visually checked.
Various modifications are made in the operation unit to the image transmitted via the image guide in which such optical fibers are bundled.

The image guide is formed by bundling tens of thousands of optical fibers, and the image propagating through the image guide shows a mesh pattern due to the boundaries of a large number of optical fibers, as shown in the photograph of FIG. 6A. FIG. 7 is a diagram in which the mesh pattern (set of circles) is emphasized for easy understanding.

Further, FIG. 6B shows each optical fiber along the line segment crossing (a) as a pixel and shows RGB for each pixel. From FIG. 6B, it can be seen that the boundary portion of the optical phipher is noise as a high frequency component.

Patent Document 1 proposes an image correction method for reducing the above-mentioned mesh pattern. That is, it has been proposed to perform blurring processing by smoothing on an input image including a mesh pattern and contour enhancement processing on the blurred image.
Specifically, in the blur processing by smoothing, the first smoothing filter calculation is performed on the input image, the median filter calculation is performed on the smoothed image obtained by the first smoothing filter calculation, and the median filter is performed. The second smoothing filter calculation is performed on the median image obtained by the calculation, and in the contour enhancement processing, the pixels constituting the blurred image are sequentially set as the processing target pixels, and are included in the contour enhancement area around the processing target pixels. The average value of the gradation of the pixels is calculated, the difference between the average value and the gradation of the processing target pixel is obtained, and the gradation of the processing target pixel is multiplied by a preset coefficient for each difference. The numerical values are added to obtain a new gradation degree of the pixel to be processed.

Patent Document 2 describes color shading correction of an endoscope device including a three-plate TV camera.
When a three-color separation prism of R, G, and B is provided behind the imaging lens as an endoscope device and the resolution is increased by correcting each color, each separation prism has an incident angle dependence on the spectral characteristics. Therefore, color shading occurs in which the color tone changes between the central part and the peripheral part of the screen. In Patent Document 2, this color shading is corrected by a correction means.

In Patent Document 3, as a device that performs correction processing such as edge enhancement and smoothing on an image, the type of the image is discriminated based on the density histogram of the image in the specified region, and the discriminated image type is supported. The content of selecting the correction operation is disclosed.

Japanese Patent No. 5789472 Japanese Unexamined Patent Publication No. 11-244230 Japanese Unexamined Patent Publication No. 8-255240

In Patent Document 1, a median filter calculation is performed on the smoothed image obtained by the first smoothing filter calculation to remove pixels that are significantly different from the surroundings in the image. In Patent Documents 2 and 3, this median is used. No filter calculation is performed. Therefore, noise may remain in Patent Documents 2 and 3.

FIG. 8A shows a two-dimensional Fourier transform image when Gaussian processing is performed, FIG. 8B shows a two-dimensional Fourier transform image when median filtering is performed, and FIG. 8C shows a two-dimensional Fourier transform image when moving average processing is performed. It is a two-dimensional Fourier transform image. (Source, Imaging Solution Aiming to understand from the introduction to image processing to basics to applications)

As can be seen from FIG. 8, the high frequency component can be accurately removed by the Gaussian processing, but in the median filter processing and the moving average processing, the weight when calculating the average value is set to the pixel of interest (center pixel) as in the Gaussian processing. Since the weight is not processed to increase the weight as it gets closer and decrease it as it gets farther, it is not possible to accurately remove the high frequency component. Therefore, although the mesh pattern due to the boundary portion of the optical fiber can be blurred, it cannot be removed.

In the image correction method according to the present invention in order to solve the above problems, the pixel values are weighted by using a Gaussian filter as a low-pass filter for an image input from an image guide in which optical fibers are bundled, thereby forming a mesh formed by the boundary of the optical fibers. The pattern was removed, and then the contour of the subject in the part of the mesh pattern removed by the smoothing process by the Gaussian filter was restored by using a LOG (low-passian of Gaussian) filter.

By using a LOG filter in which the resolution of the circular image corresponding to the outer diameter circle of the optical fiber shows the maximum value as the image restoration filter, the integrated value of the inner portion and the outer portion of the outer diameter circle of the optical fiber becomes small and is not emphasized.

According to the image correction method according to the present invention, in the smoothing process by the Gaussian filter, the weighting of the pixel center portion is increased and the weighting of the pixel peripheral portion is decreased, so that the mesh pattern formed by the adjacent portion of the optical fiber is erased. To.
Since there is no image in the part where the mesh pattern is erased, it looks blurry. Therefore, the image of the subject contour portion (boundary portion) included in the once erased optical fiber image is restored by using the LOG filter.

In the restoration, by using a two-stage LOG filter, the original network structure is not restored, but the adjacent parts of each optical fiber are corrected so as to be naturally continuous. The first-stage LOG filter is used as a coefficient suitable for the optical fiber diameter to cut the mesh pattern component included in the optical fiber image, and the second-stage LOG filter is used as the contour enhancement filter to restore the subject contour portion. Therefore, when used as an endoscope, an image with higher reliability than before can be obtained.

Diagram explaining a general fiberscope The figure explaining the modification process of this invention (A) is a fiberscope photograph after Gaussian treatment, and (b) is a graph showing the RBG of an optical fiber (pixel) along a line segment crossing the photograph (a). (A) is a fiberscope photograph that has undergone unsharp mask processing after Gaussian treatment, and (b) is a graph showing the RBG of an optical fiber (pixel) along a line segment that crosses the photograph (a). (A) is a fiberscope photograph that has undergone LOG filtering after Gaussian treatment, and (b) is a graph showing the RBG of an optical fiber (pixel) along a line segment that crosses the photograph (a). (A) is a fiberscope photograph when no processing is performed, and (b) is a graph showing the RBG of an optical fiber (pixel) along a line segment crossing the photograph (a). Diagram that emphasizes the image of a conventional fiberscope (A) is a two-dimensional Fourier transform image when Gaussian processing is performed, (b) is a two-dimensional Fourier transform image when median filter processing is performed, and (c) is a two-dimensional Fourier transform image when moving average processing is performed. Converted image

Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings. The fiberscope consists of a flexible portion 1 to be inserted into the body or the like and an operation portion 2 to be gripped by a hand. An image guide 3 in which a large number of optical fibers are bundled is provided in the flexible portion 1, and an objective lens 4 is provided at the tip of the flexible portion 1. Is attached, and a light guide fiber 5 for illumination is arranged along the image guide 3, and the light guide fiber 5 radiates the light taken in from the light source from the tip of the flexible portion 1.

The irradiation light from the tip of the flexible portion 1 hits the object, the reflected light is incident on the objective lens 4, and the focal length and the like are set so that the incident light is imaged at the tip of the image guide 3.

The light imaged at the tip of the image guide 3 is corrected by a correction unit arranged in the operation unit 2 or the like, and then observed through the eyepiece 6. In the present invention, the observed image is the one in which the mesh pattern is removed by the correction portion.

FIG. 2 is a diagram illustrating a modification process of the present invention, and the image entered in the modification section is first applied to a low-pass filter (Gaussian filter). In this low-pass filter, the neighboring pixel values are weighted using the following Gaussian distribution (Equation 1).

In the case of a Gaussian filter with a standard deviation of δ = 1.3 and a neighborhood of 8, the kernel k is, for example, a 17 × 17 determinant.

FIG. 3 (a) is a fiberscope photograph after Gaussian treatment, and FIG. 3 (b) is a graph showing an RBG of an optical fiber (pixel) along a line segment crossing the photograph (a). The boundary (noise) has disappeared.

In order to erase the boundary portion of the optical fiber, it is necessary to use a Gaussian filter having a diameter larger than the diameter of the optical fiber to increase the weight of the central portion of the pixel and reduce the weight of the peripheral portion.

In the fiberscope photograph after Gaussian treatment, the boundary of the optical fiber disappears, but since there is nothing at the boundary, the image is totally blurred. Contour enhancement and unsharpness processing have been conventionally known as means for obtaining a sharp image from such a blurred image.

FIG. 4 (a) is a fiberscope photograph in which an unsharp mask treatment is performed after Gaussian treatment, and FIG. 4 (b) is a graph showing an RBG of an optical fiber (pixel) along a line segment crossing the photograph (a). As is clear from this figure, if the unsharp mask treatment is performed after the Gaussian treatment, the noise is restored even if the sharpness is increased.

Therefore, in the present invention, a LOG filter (Laplacian of Gaussian filter) is used to restore frequencies other than the frequency components of the fiber contour (mesh pattern). The formula of this LOG filter is shown below.

The LOG filter can emphasize only the required frequencies in a limited way. That is, by using a LOG filter that most responds to the frequency of the circle corresponding to the outer diameter circle of the optical fiber (maximum resolution), only the original frequency component of the fiberscope image is selected while the noise at the boundary is removed. The image is emphasized.

By multiplying the frequency of the image after the Gaussian processing by n with the LOG filter, the frequency is pseudo-shifted to the higher side. As a result, the low-frequency image in the previous stage is restored to the original frequency (or higher).

The fiberscope to which the image correction method according to the present invention is applied can also be applied to industrial use other than medical use.

1 ... Flexible part, 2 ... Operation part, 3 ... Image guide, 4 ... Objective lens, 5 ... Light guide fiber, 6 ... Eyepiece.

Claims (1)

The image input from the image guide in which the optical fibers are bundled is weighted with a Gaussian filter as a low-pass filter to remove the mesh pattern due to the boundary of the optical fibers, and then removed by the smoothing process by the Gaussian filter. Use a two-stage filter that shows the maximum resolution of the circular image corresponding to the outer diameter circle of the optical fiber for frequencies other than the contour frequency of the mesh pattern part, and use the first-stage filter of these two-stage filters. An image correction method characterized in that a mesh pattern component is cut by using a bandpass filter having an optical fiber diameter, and a subject ring portion is restored by using a second-stage filter as a contour enhancement filter.

Images