JP2550007B2 - Defect detection device for solid-state image sensor - Google Patents

Description

The present invention relates to a defect detection apparatus for a solid-state image sensor, and is applied to a case where a quality of a product is judged when the solid-state image sensor used in a video camera is manufactured. And is suitable.

[Background technology and its problems]

Since this type of solid-state image sensor has a large number of pixels formed on a plane lattice, it is practically difficult to obtain a solid-state image sensor having uniform characteristics over the entire light-receiving surface of the image sensor during manufacturing. The photoelectric conversion output obtained from the pixels arranged in sequence contains various noise components.

That is, when the photoelectric conversion output RD is detected for one horizontal section, as shown in FIG.
RD is the shading component SH where the average signal level gradually fluctuates when the solid-state image sensor is horizontally scanned.
With. This shading component SH is generated on the basis of unevenness of aperture of each pixel, unevenness of conversion, unevenness of etching, unevenness of dyeing of a filter, etc., and has a relatively low frequency.

Secondly, the photoelectric conversion output RD has a noise component K0 generated based on the unevenness between pixels. This noise component K0 has a relatively high frequency, and its fluctuation range is also relatively large.

The above is a phenomenon that can be seen in a solid-state image sensor that does not include a defective pixel. However, if there is a defective pixel, the light-to-multiple conversion output RD is a white pulse rising at a position corresponding to this if there is a defect. A defect WD or a black defect BD will occur.

In this way, the detection data RD including various fluctuation components
Conventionally, a method of extracting the detection data RD using a spatial filter has been used in order to detect the defective pixels WD and BD based on the defective pixels from among these and detect the defective pixels. However, according to this method, it is necessary to set the size of the spatial filter and the weighting function so as to adapt to the size, type, or frequency characteristic of the defect to be detected, but actually set the optimum value. In many cases, it is difficult to detect data RD based on frequency characteristics.
It is extremely difficult in practice to design a spatial filter having an optimum frequency characteristic when detecting a defect from an image, and it is practically difficult to realize it with an easy filter such as a 3 × 3 or 5 × 5 filter. is there.

Further, according to the conventional defect detection method, when the defect is determined by comparing the output obtained from the space filter to which the detection data RD is supplied with the reference value in the comparator, the defect collection state and When the signal level of the detection data RD changes, the output level changes accordingly, so that the true data obtained from the defective pixel cannot be estimated, which makes it difficult to automatically determine the defective pixel.

 For example, the set of Laplacian -1-1-1 -1 8-1 -1-1-1 for defects changes as shown in FIG.

Further, according to the conventional defect detection method, when performing the defect detection processing using the spatial filter, it is not possible to obtain the determination result for the pixels in the area of the outer peripheral edge of the solid-state image sensor. In this connection, this conventional defect detection method generally takes a wide range of spatial filtering and performs filtering a plurality of times. In this way, it is unavoidable to generate a blank area at the periphery.

For example, when a 3 × 3 space filter is used for the defect detection data RD, the calculation cannot be performed for one bit of pixels in the outer peripheral area as shown in FIG.

[Object of the Invention]

The present invention has been made in consideration of the above points, and accurately determines a defect included in the detection data RD having a large variation component based on the shading component and the noise component with a relatively simple configuration. In order to obtain a defect detection device capable of performing such a defect detection device, a defect detection device capable of surely executing a defect detection calculation for pixels of the outer peripheral edge of the solid-state image sensor is proposed. Is what you are trying to do.

[Outline of Invention]

In order to achieve such an object, in the present invention, defect data contained in the detection data by extracting a shielding component from the detection data obtained from the solid-state imaging device and subtracting the extracted shielding component from the detection data. To obtain defect detection data.

〔Example〕

Referring to FIG. 1, reference numeral 1 denotes a solid-state image sensor, which is composed of a large number of pixels, for example, CCDs (Charge Coupled Devices) arranged on a plane lattice in the H direction and the V direction, and photoelectrically converts light emitted from a light source 2. The photoelectric conversion output of each pixel is sequentially scanned by 1H by the horizontal clock signal SH and the vertical clock signal SV generated by the clock driver 3, and is input as the time-serial detection data RD to the input circuit 4
Is input to the analog-digital conversion circuit 5 through
After being converted into a digital signal, it is input to the picture memory 7 of the shielding extraction circuit 6. Picture memory 7
Each time it receives the detection data RD that comes in every 1H, it performs a synchronous addition operation for each pixel using the accumulator 8 and thereby performs signal processing to reduce the noise component included in the detection data RD. After that, the detected data is stored and held. Thus, as shown in FIG. 2 (B), the high frequency noise component K0 based on the unevenness of each pixel included in the detection data RD (FIG. 2 (A)) is suppressed and converted into a small noise component K1. The input data S1 will be sent from the picture memory 7.

Here, as the input data S1, the defect signals WD and BD based on the abnormality of the pixel of the solid-state image sensor 1 and the shading component SH that fluctuates slowly during the 1H period are sent as they are. Here, since the shading component SH has the property of changing with a similar tendency with respect to the entire 1H line signal that sequentially arrives, it is output as it is without being suppressed by the synchronous addition for each frame. Also the defect signal WD
Since BD and BD are generated as signals similar to a plurality of adjacent line signals, they are not suppressed and remain as they are. This is because, when a CCD is used as a solid-state image sensor, anomalies that occur in pixels due to its nature do not continue in the H direction in most cases, but continue in the V direction.

The input data S1 having such a signal content is input to the defect removing filter 9. The defect removing filter 9 is composed of a 1 × 5 median filter as shown in FIG.
The intermediate value of the pixel data is transmitted as the intermediate position data. That is, in FIG. 5, the data "40" of the i-th vixel (whose address is xij) contained in the j-th line output of the input data S1 is the central bit position, that is, the third bit position. The addresses x (i-1) j, x (i-) that are input to M3 and are sequentially connected
2) The data “12” and “10” of the pixel of j are second and first.
At bit positions M2 and M1 of address x (i +
1) The data "9" and "11" of j and x (i + 2) j are input to the fourth and fifth bit positions M4 and M5. In this state, the filter 9 outputs the data representing the intermediate value of the data input to the first to fifth bit positions M1 to M5 (the data "11" of the fifth bit in this embodiment). Is output as S2.

In such an operation, the content "40" of the data input to the third bit position M3 shows a large value representing a white defect, whereas the value of the first, second, fourth and fifth bit positions is that value. Contents of M1, M2, M4, M5 "10", "12", "9",
"11" has a signal level indicating that the corresponding pixel is normal. In such a case, the median filter operation causes the data at the central position xij to change to the intermediate value data "1.
1 ”(data at address x (i + 2) j) and sends it as output S2. In this way, the defect removing filter 9 can output the filter output S2 from which the defect is removed even if the input data S1 contains a defect.

When the 1 × 5 median filter is used as the defect removing filter 9 as described above, the defect removing operation cannot be performed if there are 3 or more defects in 5 consecutive bit inputs. In this case, it is determined that the solid-state image sensor under inspection is defective.

This filter output S2 is input to the buffer memory 10,
Noise components are further suppressed by performing simultaneous addition operation using the accumulator 8 and then the shaded extraction output S3 as shown in FIG. 2 (C) is transmitted. Here, as compared with the filter output S2 (FIG. 2 (B)), the shaded extraction output S3 has the defect data WD and BD removed and the noise component K2 based on the unevenness of each pixel is further reduced. This shielding extraction output S3 is given to the low-pass filter 11.

This low-pass filter 11 has a shielding extraction output S3.
For smoothing the noise component K2 contained in, for example, a 3 × 3 low pass filter as shown in FIG. 6 or 7 can be applied. Thus, at the output end of the low-pass filter 11, as shown in FIG.
A smoothed output S4 is obtained, which is the peripheral interpolation circuit.
Given to twelve.

The peripheral edge interpolating circuit 12 cannot avoid the occurrence of a blank area which cannot be calculated by calculation in the peripheral edge portion of the solid-state image pickup device 1 during the processing in the defect removal filter 9 and the low-pass filter 11, so that the data in this blank area is inevitable. Is interpolated so that the subsequent processing can be performed. Incidentally, if there is a pixel group for which it is not actually determined whether or not the solid-state image sensor 1 can be used for the outer peripheral edge portion, the peripheral edge portion cannot perform the photoelectric conversion function, so the solid-state image sensor is effectively used. In order to do so, it is better that there is no such undecidable area.

In practice, as shown in FIG.
Uses a 5 median filter and low-pass filter 11
Assuming that a 3 × 3 filter is used, the smoothed output S4 has three bit blank areas 21 and 22 at the outer peripheral edge in the H direction and 1 at the outer peripheral edge in the V direction as shown in FIG.
Blank areas 23 and 24 of the bit result. This blank area
21 to 24 are interpolated by the peripheral interpolation circuit 12 in the order shown in FIG.

That is, first, the blank area 21 on the left edge is shown in FIG. 9 (A).
Interpolation is performed using the data D1 of the 3-bit pixel group inside the blank area 21 as shown in FIG. Thus, as shown in FIG. 9 (B), two data areas having the same data D1 are contiguously formed on the left peripheral portion.

Next, the peripheral interpolation circuit 12 interpolates the data D2 of the pixel group of one bit inside the blank area 23 of the upper edge to have the same data D2 as shown in FIG. 9C. Interpolation is performed to connect two data areas. Subsequently, as shown in FIG. 9 (C), the peripheral portion interpolation circuit 12 carries out an interpolation operation for the blank area 22 on the right side edge portion by the data D3 inside the blank area 22. As a result, as shown in FIG. 9 (D). , Two areas having the same data D3 are connected to each other.

After that, the peripheral edge interpolation circuit 12 interpolates the data D4 inside the blank area 24 at the lower edge, and thus the ninth area
As shown in FIG. 7E, two areas having the same data D4 are formed at the lower edge.

In this way, the interpolation output S5 generates an interpolation output S5 having corresponding data over the entire pixels of the solid-state image sensor 1, and the subtraction circuit 13 subtracts this from the input data S1 sent from the picture memory 7.

Here, the input data S1 is the detection data RD obtained in the solid-state image sensor 1 as described above with reference to FIG.
Since the data is the same as that when the noise component K0 in FIG. 2 (A) is suppressed, the output of pixels of the entire solid-state image sensor 1 is output including that of the peripheral portion. Therefore, the subtraction output S6 of the subtraction circuit 13 is the input data S1 (FIG. 2 (B)).
Then, by subtracting the shading component SH from the interpolation output S5, only the white defect WD and the black defect BD are output.

Since such subtraction is performed over the entire surface of the solid-state image sensor 1 including the outer peripheral edge portion, if the white defect WD or the black defect BD occurs in the peripheral pixel, it can be reliably detected. By the way, the same data as the data of the pixel group inside is interpolated in the signal component corresponding to the outer peripheral portion of the interpolation output S5, but in reality, the shading curve
Since a curve similar to the curve at the outer edge of SH can be obtained, it is possible not to leave the influence of the shaded waveform on the subtraction result with the input data S1.

The subtraction output S6 of the subtraction circuit 13 is input to the defect determination circuit 14.

When the defect judgment circuit 14 receives the subtraction output S6, FIG.
For example, four comparison levels COM1 to COM4 are set, and it is determined whether the defect WD or BD exceeds these comparison levels COM1 to COM4. And according to the judgment result,
First, so-called score calculation is performed by integrating the number of defects, and the score of the currently inspected solid-state image sensor 1 is evaluated.

Secondly, the defect determination circuit 14 detects the address where the defect WD or BD occurs and performs so-called shape recognition, score calculation, and post-processing.

The third defect judgment circuit 14 creates a defect map representing the distribution and size of defects based on the shape recognition and the score calculation.

Thus, the defect judgment circuit 14 sends out the defect detection signal ADD containing the result of the defect judgment.

With the above configuration, when the detection data RD as shown in FIG. 10 (A) is obtained from the solid-state image sensor 1, the filter output S2 is shown in FIG. 10 by the operation of the median filter in the defect removal filter 9. As shown in (B), there is only the shielding component SH, and this is the low-pass filter.
After the smoothing in 11, the interpolation calculation is performed in the rear edge interpolation circuit 12. When the subtraction circuit 13 subtracts from the input data S1 using the interpolation output S5, the subtraction output S6 is the same as the defect signals WD and BD separated and extracted from the shaded component SH as shown in FIG. 10C. You can get a signal. Therefore, in the defect judgment circuit 14, the defect WD and
The BD address can be easily identified.

The defect removing filter 9 for a color image may have the structure shown in FIG. That is, as shown in FIG. 11 (A), the solid-state image sensor 1 has three primary color signals R,
Although it has a configuration in which pixels corresponding to G and B are sequentially arranged, a pixel for each of the three primary color signals is provided with a unique color filter, so the defect removing filter 9
It is necessary to perform the filtering for each color. In such a case, for each of the primary color signals R, G and B, as shown in FIG.
The primary color signals may be separated from each other by using the mask 21 formed by SK, and the separated output may be applied to the median filter.

Here, the filter used in the present invention will be described with reference to FIG. By comparing adjacent two-bit data among the input data S1 extracted from the picture memory 7 and selecting the maximum value or the minimum value thereof, a filter containing no black defect or white defect is selected. It tries to get the output.

In this embodiment, the defect removing filter 9 is the pixel xi.
To obtain the filter output S2 of j, the data of five adjacent bits before and after are used, and two adjacent bits are used in the first step.
Select the maximum value of the two bit data, select the minimum value of the two adjacent data in the second step for the selected output, and select the third value for the selected output.
In step 4), the minimum value of the two adjacent data is selected, and for the selected output, the maximum value of the two data is selected in the fourth step. In this way, the filter output can be obtained by removing the data due to the defect based on the defective pixel as the data at the pixel xij position. For this reason, the calculation of the median filter shown in FIG. 5 takes too much time and can be used as a simple configuration as an alternative to a device having a problem that it cannot be applied in hardware.

Further, instead of the defect removing filter 9 described above, a structure using a space filter or a two-dimensional filter may be applied.

When processing a color image as the low-pass filter 11 having the configuration shown in FIG. 1, the one having the configuration shown in FIG. 13 can be applied. That is, as shown in FIG. 13, the shading output S3 received from the buffer memory 10 is stored in the memory area M11 as the center for storing data on, for example, the R signal. Next, the memory M11 is shifted upward by 1 bit to take in the lower data R3, and then the memory M11 is shifted right by 3 bits to take in the data R4.

Then, the low-pass filter 11 obtains the averaged output MW using the following equation (1) MW = (R1 + R2 + R3 + R4) / 4 (1) and sends this as the smoothed output S4.

By using the low-pass filter 11 that performs signal processing in this manner, it is possible to realize a low-pass filter that can be applied to image signals of any color pattern and that can simultaneously process all pixels.

〔The invention's effect〕

As described above, according to the present invention, by performing a process of selectively outputting one detection data having an intermediate value between the detection data corresponding to a predetermined number of pixel groups which successively follow each pixel of the detection data. Utilizing the conventional spatial filter by extracting the shading component contained in the detection data and by subtracting the extracted shading component from the detection data, the defect component contained in the detection data can be extracted. As in the case of directly extracting the defect component contained in the detection data from the detection data, it is necessary to perform the extremely difficult task of setting the size of the spatial filter, the weighting coefficient, and the frequency characteristic to the optimum values. By repeating simple processing, the system is not affected by noise and defects contained in the detected data. It can be reliably detected by only a simple configuration Deingu component. Therefore, even in the defect detection by subtracting the shading component, it is not necessary to use the uncertainty element of the prediction coefficient setting, and the defect can be detected with a simple configuration and high accuracy.

Further, according to the present invention, even when the signal level of the detection data is changed, the defect detection data indicating the defect component is obtained by performing processing such as subtracting the aging component from the detection data and comparing with a predetermined level. By doing so, the true defective output data can be obtained by removing the noise due to the pixel unevenness.

[Brief description of drawings]

FIG. 1 is a block diagram showing an entire defect detecting device for a solid-state image pickup device used in the present invention, FIG. 2 is a signal waveform diagram showing signals of respective parts thereof, and FIG. 3 is an explanation of a spatial filter used conventionally. FIG. 4 is a schematic diagram showing a blank area produced when a space filter is used, FIG. 5 is a schematic diagram showing the defect removing filter 9 of FIG. 1, FIG. 6 and FIG. A chart showing the low-pass filter 11 of FIG. 1, FIGS. 8 and 9 are schematic diagrams for explaining the interpolation operation of the peripheral portion interpolation circuit 12 of FIG. 1, and FIG. 10 is a defect due to the configuration of FIG. Signal waveform diagram of each part showing the experimental results of the detector, FIG. 11 and 12
1 is a schematic diagram showing the present invention of the defect removing filter 9 shown in FIG. 1, and FIG. 13 is a schematic diagram showing another embodiment of the low-pass filter 11 shown in FIG. DESCRIPTION OF SYMBOLS 1 ... Solid-state image sensor, 2 ... Light source, 3 ... Clock driver, 6 ... Shading extraction circuit, 7 ... Picture memory, 8 ... Akymulator, 9 ... Defect removal filter, 10 ... Buffer memory, 11 ...... Low-pass filter, 12 ...... peripheral part interpolation circuit, 13 …… subtraction circuit, 14 …… defect judgment circuit.

 ─────────────────────────────────────────────────── ─── Continuation of the front page (72) Inventor Akio Kanemaru 6-735 Kitashinagawa, Shinagawa-ku, Tokyo Within Sony Corporation (56) References JP-A-52-100829 (JP, A) JP-A-SHO 56-111383 (JP, A)

Claims (1)

(57) [Claims] 1. An AD converter for converting first detection data obtained from a solid-state image sensor into a digital signal, and a predetermined number of pixels successively receiving the second detection data AD-converted by the AD converter. Two pieces of the second detection data corresponding to the group are sequentially used, the maximum value of the two bit data is selectively output in the first step, and the maximum output of the two bits is selected in the second step. Each of them is sequentially used to selectively output the maximum value of the two data, and in the third step, two of the selected outputs are sequentially used to selectively output the minimum value of the two data, and the fourth step is selected. 2 of the selected outputs are sequentially used in
From the second detection data, a subtraction circuit that subtracts the extracted shading component from the second detection data, and a subtraction circuit that subtracts the extracted shading component from the second detection data. And a defect determination circuit that outputs defect detection data indicating defect data by comparing the subtraction output of the above with a predetermined level.