JPH1118028A - Shot changeover detection method and recording medium recording shot changeover detection program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a method that detects a shot changeover with a long changeover time with which the shot changeover including dissolve is conducted stably at a high speed, even if a camera or an object moves in the method, that detects shot changeover from video data. SOLUTION: A label Lc denoting that a pixel value takes almost a prescribed value for each pixel point, a label Ls denoting that the pixel value has suddenly changed, and a label Ll denoting that the pixel value has changed almost linearly for a time period are given based on timewise fluctuation of the pixel value to the pixel value (S3). The presence of a cut is discriminated, when an incident ratio of the label Ls calculated from pixel points other than those without the provision of the labels is high (S51, S52), and the presence of dissolve or fade is discriminated when an incident ratio of the label L1 calculated from pixel points other than those without provision of the labels is high (S53, S54). Furthermore, a wide/narrow time interval of input of an image string is controlled by the magnitude of the motion of the video image.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a shot switching detection method for detecting shot switching from video data and a recording medium on which a shot switching detection program is recorded. When constructing a video database, it can be used for processing of dividing video into scenes.

[0002]

2. Description of the Related Art An image is composed of a plurality of shots (scenes continuously shot by one camera). A shot-shot break is called “shot switching”. There are various methods for switching shots.

[0003] The most common method is called "cutting", and as shown in FIG. 10 (A), instantaneously switches from one shot to the next shot. Techniques for slowly switching shots over time include "dissolve", "fade in", "fade out",
"Wipe" and the like. Dissolve is a method of switching from a certain shot to the next shot while fusing each other. FIG. 10B shows an example of the dissolve.
Fade-in is a method of switching from a black (or single color) image so that the next shot gradually emerges. Fade out is fade out
This is the reverse operation of IN, and is a method of switching to a black (or single color) image while gradually thinning a certain shot. Wipe is a technique for gradually displaying the next shot from a part of the screen when switching from one shot to the next shot.

Various methods have been proposed as methods for detecting shot switching from video data. For convenience of description, three adjacent images in the video are represented by F0, F1, and F2.
And In order to detect a cut in the shot switching, it is sufficient to detect that the patterns of the two images F0 and F1 are significantly different.

[0005] Various methods have been proposed as a method for evaluating the difference between the pictures of two images. As one method,
There is one that calculates a difference for each pixel between images and determines that there is a cut when the sum of the differences is large.
No. 4273, “Index image creation device”). In addition, there is a method in which a color histogram is calculated for each image, and a difference between the color histograms is calculated to detect a cut (Japanese Patent Laid-Open No. 4-111181, "Moving image change point detection method").

A conventional technique for detecting dissolve and fade in shot switching will be described (reference: A. Ha).
mpapur, R. Jain, and T. Weymouth: Digital Video Segm
entation. Proceedings of ACM Multimedia 94, ACM Pr
ess, pp. 357-364, 1994).

In order to detect dissolves and fades, the property that pixel values monotonically increase (or decrease) in the process of dissolves and fades is used. The pixel values of the corresponding pixels of the image rows F0, F1, and F2 are f0, f1, and f2.

From the relationship that the pixel value monotonically increases (or decreases), f2−f0 ≒ 2 (f2−f1) holds (where ≒ indicates that the values are almost equal). That is, the relationship (f2-f1) / (f2-f0) ≒ 1/2 [1] holds for all pixels. Image F0,
For each pixel of F1 and F2, (f2-f1) / (f2-
A new image Ic having a pixel value of f0) is created. From equation [1], ideally, all the pixel values of Ic should be 』[2]. [2] is not strictly established because the actual image data contains noise or motion of the subject, but approximately [2]
Is assumed to hold. In order to determine whether or not the image sequences F0, F1, and F2 include a dissolve, it is only necessary to determine whether [2] is approximately established. In order to determine whether or not [2] holds, constancy = N (Id) / ｛σ (Ic) · (1.0+ | Cx−cx |
+ | Cy−cy |)}, and when the function value exceeds a certain threshold value, it is determined that there is a dissolve. Take). Where N
(Id) is the number of pixels that satisfies f2-f1，0, and σ (Ic)
Is the standard deviation of the pixel values of the image Ic, (Cx, Cy) is the center of gravity calculated using the luminance of the image Ic as a weight, and (cx, cy)
) Is the center of the image.

[0009] The method described above uses three images F0, F1,
From F2, the presence or absence of cut and dissolve was determined. In order to enumerate all the shot changes in the video data, the above-described determination method may be repeatedly applied while sequentially extracting images from the video. In the conventional method, a method is used in which an image is extracted from a video at regular time intervals and processed.

[0010]

The above-mentioned prior art has the following three problems. (1) Detection errors easily occur when the camera or subject moves significantly. This is because N (Id) increases when the movement of the camera or the subject is included, and thus
Is increased.

(2) Since the method of sequentially examining images while extracting them from the video data at regular time intervals is employed, there is a problem that it takes a long processing time. In a video, scenes with little change (such as scenes of scenery) and scenes with sharp changes (such as action scenes) are often mixed.
In a scene with little change, the performance does not decrease even if the sampling interval is widened. However, if the sampling interval is widened in a scene with rapid change, the detection error increases due to the problem (1) above.
In order to reduce errors, it was necessary to extract and process images at small intervals.

(3) There is a problem that very slow shot switching (shot switching with a long switching time) cannot be detected. In a very slow shot change, the image sequences F0, F1,
No significant change is observed between F2. Therefore, in the equation [1], the denominator of (f2-f1) / (f2-f0) becomes small, and the variation of the value, that is, the standard deviation σ (I
c) becomes large, and the constancy takes a small value, so that there is a problem that the dissolve cannot be detected.

SUMMARY OF THE INVENTION The present invention has been made to solve the above problems. (1) Even if there is a movement of a camera or a subject, a shot change including a dissolve can be detected stably. High-speed processing can be performed while adjusting the sampling interval adaptively according to the degree of change involved. (3) Shot switching detection method and shot switching detection program that can detect shot switching with long switching times It is an object of the present invention to provide a recording medium that has been used.

[0014]

According to a first aspect of the present invention, there is provided a labeling process for assigning a label to each pixel point based on a time variation of a pixel value, and a shot switching based on an appearance ratio of the label. And a label Lc indicating that the pixel value takes a substantially constant value as a label given in the labeling process.
And a label Ls indicating that the pixel value has suddenly changed,
In the process of the shot switching determination, it is characterized in that it is determined that the shot switching exists when the appearance ratio of the label Ls calculated excluding the pixel points to which no label is added is large.

The "label appearance ratio" is calculated excluding "pixel points to which no label is given". “Pixel points without labels” are points at which pixel values fluctuate irregularly due to the movement of a camera or a subject, and are a main cause of detection errors in the prior art. . Since the “label appearance ratio” is calculated excluding such points, the “label appearance ratio” is less likely to be adversely affected by the movement of the camera or the subject. That is, by determining the presence or absence of a shot switch based on the “label appearance ratio”,
Shot switching can be detected stably even when the camera or the subject moves. This solves the problem (1).

According to a second aspect of the present invention, a label L1 indicating that the pixel value has changed substantially linearly in a certain time interval is included as a label assigned in the labeling process. When the appearance ratio of is large, it is determined that there is a dissolve or a fade.

By determining the presence or absence of a shot switch based on the "appearance ratio of label L1" calculated excluding "pixel points to which no label is added", dissolve and fade can be performed even when the camera or the subject moves. Shot switching including this can be detected stably. This causes problems
(1) is resolved.

According to a third aspect of the present invention, there is provided a process of inputting an image sequence by designating a time and a time interval, and narrowing the time interval when a large motion is included in the video so that the motion included in the video is reduced. A step width adjustment process for controlling the time interval to be extended when the number is small.

As described above, since the detection process is performed while adjusting the thinning interval adaptively according to the degree of motion included in the video, the trouble of processing unnecessary images can be omitted, and the process can be performed at high speed. it can. Further, by extending the time interval, a dissolve or a fade having a long switching time can be detected. This solves problems (2) and (3).

According to a fourth aspect of the present invention, a large motion is included in a video when the number of pixel points to which no label is added is large, and a small motion is included in the video when the number of labels Lc is large. The time interval is controlled.

"Number of unlabeled pixels"
Is adjusted to reflect the amount of “movement” included in the video, and when the number is large, it indicates that the scene is intense. Conversely, when the number of labels Lc is large, it is a scene with little motion, so that the time interval is adjusted to be widened. As described above, since the detection processing is performed while adaptively adjusting the thinning-out interval according to the degree of the movement included in the video, the trouble of processing unnecessary images can be omitted, and the processing can be performed at high speed. Further, by extending the time interval, a dissolve or a fade having a long switching time can be detected. As a result, problems (2) and (3)
Is resolved.

According to a fifth aspect of the present invention, there is provided a labeling procedure for assigning a label to each pixel point based on a time variation of a pixel value, and a shot switching determining procedure for detecting a shot switching based on the appearance ratio of the label. The label switching procedure includes a label Lc indicating that the pixel value takes a substantially constant value, and a label Ls indicating that the pixel value has changed suddenly. A shot switch detection program for causing a computer to perform a process of determining that a shot switch exists when the appearance ratio of the label Ls calculated excluding the pixel points that are not performed is large is recorded on a computer-readable recording medium. It is characterized by having done.

Thus, the shot change detection method according to the first aspect can be executed by a computer, and the problem (1) can be solved. The invention according to claim 6 further includes, as a label assigned by the labeling procedure, a label Ll indicating that the pixel value has changed substantially linearly in a certain time interval, and the shot switching determination procedure includes a step of determining a pixel point to which no label is assigned. Label L calculated excluding
A shot switching detection program for causing a computer to perform a process of determining that a dissolve or a fade exists when the appearance ratio of 1 is large is recorded on a computer-readable recording medium.

Thus, the shot change detection method according to claim 2 can be executed by a computer, and the problem (1) can be solved. According to a seventh aspect of the present invention, there is provided a procedure for inputting an image sequence by designating a time and a time interval, and narrowing the time interval when a large motion is included in the video, and setting the time interval when the motion included in the video is small. A shot switching detection program having a step width adjustment procedure for controlling so as to increase the time interval is recorded on a computer-readable recording medium.

Thus, the shot change detection method according to the third aspect can be executed by a computer, and the problems (2) and (3) can be solved. The invention according to claim 8 is characterized in that, as the step width adjustment procedure, a large motion is included in the video when the number of unlabeled pixel points is large, and is included in the video when the number of labels Lc is large. The shot change detection program for controlling the time interval is recorded on a computer-readable recording medium assuming that the movement to be performed is small.

Thus, the shot change detection method of claim 4 can be executed by a computer, and the problems (2) and (3) can be solved.

[0027]

DESCRIPTION OF THE PREFERRED EMBODIMENTS One embodiment of the present invention will be described with reference to the drawings. (1) Outline of Processing An outline of processing will be described with reference to the flowchart of FIG.

First, t is a variable for storing the frame number of the image currently focused on, and s is a variable for storing the step width (time interval for extracting an image from a video). Substitute t = 0 and s = 4 as initial values (step S1).

Next, as shown in FIG. 2, four images F0, F1, F2 and F3 are extracted from the t-th frame at an interval s (step S2). Subsequently, a “labeling procedure” described later is called, and as a result, the number N of each label is obtained.
c, Ns, Nl, and the number of pixel points Nu to which no label is given are obtained (step S3).

Next, a "step width adjustment procedure" described later.
Is called, and it is checked whether or not the step width s is appropriate. If it is determined that adjustment is necessary, the process returns to step S2 with the updated step width (step S4).

Next, a shot switching determination procedure is performed (step S5). Here, the appearance ratio Ns /
If (N-Nu) is greater than the threshold value Tcut, it is determined that "cut exists" (steps S51 to S52). If the appearance ratio Nl / (N-Nu) of the label Ll is larger than the threshold value Tdiss, it is determined that "dissolve (or fade) is present" (steps S53 to S54).

Next, it is checked whether or not the end of the image has been reached (step S6). If the end has not been reached, t is advanced by the step width s (step S7), and step S2 is performed.
Return to the processing of.

(2) Labeling Procedure (FIG. 1: Step S3) The labeling procedure will be described with reference to FIG. FIG.
9 is a processing flowchart of a labeling procedure.

First, coordinate values (x,
y) is initialized to (0, 0) (step S301).
As shown in FIG. 4, an image F0, corresponding to the point (x, y)
The pixel values of F1, F2, and F3 are f0, f1, f2, and f3, and a vector obtained by combining them is f = (f0, f1,
f2, f3) (step S302). For convenience, the function var (f0, f1, f2, f3) is defined as var (f0, f1, f2, f3) = max (f0, f1, f2, f3) −min (f
0, f1, f2, f3). var () is a function that returns the fluctuation range of the value given as an argument.

In steps S303 to S308, it is checked in this order whether the following three conditions (a) to (c) are satisfied, and a label is assigned to each pixel point. (a) constant (Lc): When the vector f takes a substantially constant value, it is checked whether or not var (f0, f1, f2, f3) <T0 is satisfied (step S303). (constant label) (step S304). As schematically shown in FIG. 5A, when the pixel value takes a substantially constant value without largely changing, the Lc label is added.

(B) step (Ls): When the vector f shows an abrupt change at a certain time and has a substantially constant value before and after that, var (f0, f1) <T0 and var (f2, f3) <T0 and ｜ f0−
It is checked whether or not f3 |> T1 holds (step S305), and if it holds, an Ls label (step label) is assigned (step S306). As schematically shown in FIG.
When the pixel value indicates a stepwise change, an Ls label is given.

(C) linear (Ll): When the vector f linearly increases or decreases var (f0, f1-d, f2-2d, f3-3d) <T2 and | f3-f0
|> T1 (However, d = (f3-f0) / 3 indicates the slope of the change.)
Is checked (step S307), and if it is satisfied, an L1 label (linear label) is assigned (step S308). As schematically shown in FIG. 5C, when the pixel value linearly increases (or decreases), the L1 label is assigned.

Pixel points that do not satisfy any of the above three conditions are set to "no label" (step S30).
9). As shown in FIG. 5D, when the pixel value fluctuates irregularly, it is determined that there is no label.

The processing described above is repeated for all the pixel points. (X, y) is moved to the next pixel point (step S310). If all pixel points in the image have not been checked, the process returns to step S302, and if all pixel points have been checked, the labeling procedure ends. (Step S311).

The numbers of the pixel points to which the labels Lc, Ls, and Ll are assigned are respectively Nc, Ns, and Nl, and the number of the pixel points to which no labels are assigned is Nu. Assuming that the number of pixels of the entire screen is N, N = Nc + Ns + Nl + Nu
Holds.

The labeling method is not limited to the above embodiment. In the above embodiment, labeling is performed based on four pieces of image data, but any number of three or more pieces may be used. When the above condition is applied to the M + 1 image sequence, the above condition is generalized as follows: (a) constant: var (f0, f1, ..., fM) <T0, (b) step (k): var (f0, f1, ..., fk) <T0, and var (fk + 1, fk + 2, ..., fM) <T0, and | f0−fM |>
T1, (c) linear (i, j): var (f0, f1, ..., fi) <T0, and va
r (fj, fj + 1, ..., fM) <T0, and | f0−fM |> T1, and
var (fi, f (i + 1) -d, f (i + 2) -2d, ..., fj- (ji) d) <T2
(However, it can be labeled using d = (fj-fi) / (ji)).

The difference from the above-mentioned conditions is that the step label takes a parameter k representing the step position and parameters i and j representing the section where the linear label changes linearly. In order to determine whether or not a cut exists at the position of M / 2 in the image sequence in accordance with this change, Ns
In order to determine whether a dissolve exists by using the number of labels of step (M / 2) as
(I, j), the maximum value of the number of labels i <M / 2 <j can be used.

Further, the image is divided into blocks, and the above-described labeling procedure is executed by replacing “block” with “pixel point” and “statistical value such as average value of pixels in the block” with “pixel value”. Is also possible.

In the present embodiment, labeling is performed by simple threshold processing, but a plurality of statistical models are set, and among the set statistical models, which model most efficiently describes the time variation of the vector f. Statistical test method, MDL (Minimum Description Length)
It is also possible to use a method, AIC (Akaike Information Criteria) method or the like to determine and perform labeling. Also, labels other than those described above can be added. For example, a change in which the pixel value increases (or decreases) momentarily and then returns to the original value can be added as a new label.

(3) Shot Switching Determination Procedure (FIG. 1: Step S5) Since the pixel values change suddenly before and after the cut, the images F1 and F2
If there is a cut between, the Ls (step) label increases. By utilizing this property, the appearance ratio of the label Ls [Equation A] When CutMeasure = Ns / (Nc + Ns + Nl) = Ns / (N-Nu) exceeds a certain threshold Tcut, it is determined that there is a cut. judge.

FIG. 6 shows an example of a graph of the change over time of the cut evaluation amount (CutMeasure). The time exceeding the threshold value Tcut is output as a cut. Here, it should be noted that the ratio is obtained after subtracting the number of pixel points Nu to which no label is given. Pixel points to which no label is attached are often points where pixel values are irregularly changed due to movement of a subject or a camera, and such points cause shot switching detection errors in the conventional method. Was. Since CutMeasure is calculated excluding such a cause, it has a feature that it is hardly affected by movement.

In the dissolve, the scenes are switched in such a way that the scenes are melted. Therefore, if the images F0, F1, F2, and F3 contain the dissolve, the L1 (linear) label increases. Utilizing this property, the appearance ratio of Ll label [Equation B] DissolveMeasure = Nl / (Nc + Ns + Nl) = Nl / (N-Nu) If it exceeds a certain threshold, it is determined that dissolve exists. .

FIG. 7 shows a dissolve evaluation amount (DissolveMeas).
ure) shows an example of a time change graph. Threshold Tdiss
Is determined to be a dissolve in a section exceeding. Fade-in,
Similarly, fade-out can be detected because DissolveMeasure takes a high value. As indicated by reference numeral 701, since DissolveMeasure rarely falls below the threshold value during the dissolve, another threshold value T in addition to Tdiss is used.
diss' (set lower than Tdiss)
It is also preferable to end the dissolve when DissolveMeasure becomes lower.

In the calculation of CutMeasure and DissolveMeasure, in order to prevent the denominator from approaching zero, it is also preferable that the denominator be max (N-Nu, 0.8 * N) instead of N-Nu. .

(4) Step width adjustment procedure (FIG. 1: step S4) The step width adjustment procedure will be described with reference to FIG.
FIG. 8 is a processing flowchart of the step width adjustment procedure.

The values of Nu and Nc returned in the labeling procedure are given as inputs. If the number Nu of the pixel points to which no label is given is larger than a certain threshold T3, it is determined that the step width is too large (step S40).
1) The step width is halved (step S402).
However, the minimum value of the step width is set to 1 here.

Next, the number Nc of Lc labels is equal to the threshold T4.
If it is larger, it is determined that the step width is too small (step S403), and the step width is doubled (step S404). However, the maximum value of the step width is 16 here.

When the above two conditions are not satisfied,
Assuming that the step width is appropriate (step S40)
5), end the procedure. In a video, scenes with little movement, such as landscapes, and scenes with strong movements, such as actions, are mixed. It is necessary to reduce the step width in a scene where movement is severe. This is because Nu tends to increase, and when Nu increases, the denominator of Equations [A] and [B] becomes smaller, so that Cu is mixed only with minute noise.
This is because there is a problem that the values of tMeasure and DissolveMeasure fluctuate greatly and detection errors increase. Conversely, in situations where there is little movement, it is more efficient to increase the step width.

If the step width is small, a dissolve with a long switching time (switching very slowly) cannot be detected.
As shown in FIG. 9 (A), when the speed of the dissolve change is high, the change gradient is large, so that the change can be detected even if the step width is made small. However, as shown in FIG. Is slow,
Since the inclination of the change in the pixel value is small, the change cannot be detected. That is, it cannot be distinguished from a change caused by noise. To detect a dissolve with a long switching time, it is necessary to increase the step width.

Based on the above considerations, in this embodiment, N
When u is larger than a certain threshold value, the step width is changed to て, considering that the motion included in the video is intense. This control has an effect of preventing a detection error in a scene where movement is severe. Conversely, when Nc is larger than a certain threshold value, it is considered that the motion included in the video is small, and the step width is doubled. With this control,
There is the effect that the trouble of processing useless images can be omitted while detecting errors in a scene of rapid movement, and a dissolve with a long switching time can be detected.

Note that the step width adjustment procedure is not limited to the above embodiment. Deformation is possible within a range that does not change the purpose of adjusting the step width by reflecting the degree of “movement”. For example, the number N of Lc (constant) labels
Instead of c, the step width may be increased when (N-Nu) is larger than a certain threshold. In the above-described embodiment, the step width is changed by a power of 2, such as 1, 2, 4, 8,..., But is increased or decreased by 1, such as 1, 2, 3, 4,. Alternatively, the step width may be determined by a function set in advance.

The above-described shot switching detection method is realized using a computer program, and the program can be provided by being recorded on an appropriate recording medium such as a floppy disk, a compact disk, or a tape.

[0058]

According to the first aspect of the present invention, the "label appearance ratio" is calculated by excluding "pixel points to which no label is given" which causes a detection error, and the presence or absence of shot switching is determined based on this ratio. Since the determination is made, there is an effect that the shot change can be stably detected even when the camera or the subject moves.

According to a second aspect of the present invention, the labeling procedure includes, as a label, a label Ll indicating that the pixel value has changed substantially linearly in a certain time interval. Since it is determined that there is a dissolve or a fade, there is an effect that shot switching including the dissolve can be stably detected even when the camera or the subject moves.

According to the third aspect of the present invention, the time interval is controlled in accordance with the amount of motion included in the video, so that it is not necessary to process unnecessary images, so that high-speed processing can be performed. There is an effect that a long-time shot change can be detected.

According to the fourth aspect of the present invention, the control is performed so that the time interval is narrowed when the number of unlabeled pixels is large, so that there is an effect of reducing detection errors in a scene including a large motion. Furthermore, since the control is performed so as to increase the time interval when the number of pixel points to which the label Lc is added is large, the trouble of processing unnecessary images can be omitted, the processing can be performed at high speed, and the switching time is long. There is an effect that a shot change can be detected.

According to the fifth to eighth aspects of the present invention, the computer can execute the method of the first to fourth aspects by causing a computer to execute a shot switching detection program recorded on a recording medium. effective.

[Brief description of the drawings]

FIG. 1 is a flowchart for explaining an embodiment of the present invention.

FIG. 2 is a schematic diagram for explaining how to extract an image from a video.

FIG. 3 is a processing flowchart of a labeling procedure.

FIG. 4 is a diagram for explaining the relationship between pixel points and pixel values.

FIG. 5 is a schematic diagram for explaining a relationship between a temporal change of a pixel value and a label.

FIG. 6 is a diagram showing a change over time of a cut evaluation amount.

FIG. 7 is a diagram showing a temporal change of a dissolve evaluation amount.

FIG. 8 is a processing flowchart of a step width adjustment procedure.

FIG. 9 is a diagram for explaining an influence of a dissolve change speed.

FIG. 10 is a schematic diagram for explaining cutting and dissolving.

[Explanation of symbols]

 S1 to S7 Processing steps of the embodiment

Claims (8)

[Claims] 1. A method for detecting shot switching from an image sequence composed of three or more images, comprising: a labeling process for assigning a label to each pixel point based on a temporal variation of a pixel value; A shot switching determination process for detecting shot switching based on the label appearance ratio;
The label given in the labeling process includes a label indicating that the pixel value takes a substantially constant value and a label indicating that the pixel value has changed suddenly. A shot switching detection method, characterized in that it is determined that shot switching is present when the appearance ratio of a label indicating that the pixel value calculated excluding the pixel point that has not been given has changed abruptly is large. 2. The shot change detection method according to claim 1, wherein the label added in the labeling process includes a label indicating that a pixel value has changed substantially linearly in a certain time interval, and the shot change determination is performed. In the processing step, the dissolve or fade is determined to be present when the appearance ratio of the label indicating that the pixel value calculated excluding the pixel point to which no label has been added changes substantially linearly is large. Shot switching detection method. 3. A shot switching detection method according to claim 1, wherein a process of inputting an image sequence by designating a time and a time interval is performed, and said process is performed when a large motion is included in an image. A step of adjusting the step width to reduce the interval and increase the time interval when the amount of motion included in the video is small. 4. The shot switching detection method according to claim 3, wherein in the step width adjustment processing, a large motion is included in the image when the number of unlabeled pixel points is large. A shot switching detection method, characterized in that when the number of labels indicating that the pixel value takes a substantially constant value is large, the movement included in the video is considered to be small and the time interval is controlled. 5. A recording medium on which a shot switching detection program for detecting a shot switching from a sequence of three or more images by a computer by a computer is recorded. A labeling procedure for assigning a label to each label, and a shot switching determination procedure for detecting shot switching based on the appearance ratio of the label, wherein the label assigned by the labeling procedure has a substantially constant pixel value. A label indicating that
And a label indicating that the pixel value has suddenly changed. The shot switching determination procedure includes a step of determining when the appearance ratio of the label indicating that the pixel value has suddenly changed, excluding the pixel point to which no label has been added, is large. And recording a shot switching detection program. 6. A recording medium on which the shot switching detection program according to claim 5 is recorded, wherein the label given by the labeling procedure includes a label indicating that a pixel value has changed substantially linearly in a certain time interval. The shot switching determination procedure is to determine that a dissolve or a fade is present when the appearance ratio of a label indicating that the pixel value calculated except for the pixel point to which no label is added has changed substantially linearly is large. A recording medium on which a featured shot switching detection program is recorded. 7. A recording medium on which the shot switching detection program according to claim 5 is recorded, wherein a procedure for inputting an image sequence by specifying a time and a time interval and a large motion is included in the video. And a step width adjustment procedure for controlling the time interval to be narrowed and the time interval to be widened when the amount of movement included in the video is small. 8. The recording medium on which the shot switching detection program according to claim 7 is recorded, wherein the step width adjustment procedure includes a step in which a large motion is included in an image when the number of unlabeled pixel points is large. And controlling the time interval on the assumption that the motion contained in the video is small when the number of labels indicating that the pixel value takes a substantially constant value is small. Recording medium on which is recorded.