CN100469153C - Testing method of switching video scenes - Google Patents

Abstract

The method includes steps: first, preparing histogram statistics for mismatched pixels of moving forwards compensation block in current input image; next, calculating statistic of heterotactic characters of histogram for mismatched pixels of moving forwards compensation block between adjacent frames; then, carrying out determining procedure for switching scenes based on the said statistic, and outputting the determined result. Based on difference detection technique of moving compensation, the invention overcomes disadvantages: sensitivity to noise and movement caused by the difference algorithm for pixels located at same positions in adjacent frames; and negligence of space information of video object occurred in conventional decision method of analyzing statistical histogram of pixel intensities. Advantages are: lowering probability of false alarm and probability failed to report in detection of switching video scenes as well as raising precision and correctness of determination.

Description

Switching detection method of video scene

Technical field

The present invention relates to a kind of switching detection method of video scene, relate in particular to a kind of switching detection method of video scene that is used to promote accuracy of detection and accuracy, reduction false alarm rate and rate of failing to report in video database information indexing device and the digital video signal processor (as field/frame-rate conversion device, deinterlacing transducer) that is applicable to.

Background technology

Scene is switched, refer to the visual point of interruption in one section continuous film, the interruption of video camera or interrupting a little of film editor afterwards when it comes from film shooting usually, a film can be regarded the set of many scenes usually as, and scene then is the least unit of a film of retrieval.The video scene change detection then is to handle a most basic treatment step in the application in all video image analysis.

Because many related algorithms have appearred in the importance of scene change detection at present, wherein relatively typical prior art comprises: pixel difference algorithm and pixel histogramming algorithm.

Because scene is switched the cutting on the time dimension that can be regarded as video sequence in essence, it shows the most intuitively and is, has violent saltus step between scene switching point consecutive frame coordination pixel.Existing pixel difference algorithm (detecting the method and apparatus of scene change and adjusting picture coding type in a kind of high definition television encoder as Chinese No. 99813804 patent application announcement) is exactly to differentiate the scene switching by the saltus step statistic between the coordination pixel between the detection consecutive frame.Saltus step statistic commonly used has coordination pixel mean absolute deviation MAD (seeing also formula 1) and coordination pixel mutation probability statistics DP (seeing also formula 2).

$\mathrm{MAD}=\frac{1}{H\×W}\underset{\stackrel{\mathrm{\ρ}}{X}\∈I}{\mathrm{\Σ}}|f_1\left(\stackrel{\mathrm{\ρ}}{X}\right)-f_2\left(\stackrel{\mathrm{\ρ}}{X}\right)|---\left(1\right)$

$\mathrm{DP}=\frac{1}{H\×W}\underset{\stackrel{\mathrm{\ρ}}{X}\∈I}{\mathrm{\Σ}}(|f_1\left(\stackrel{\mathrm{\ρ}}{X}\right)-f_2\left(\stackrel{\mathrm{\ρ}}{X}\right)|>T?1:0)---\left(2\right)$

F in the above-mentioned formula ₁And f ₂Represent the pixel value of present frame and former frame respectively, H and W then distinguish the horizontal direction of representing images and the pixel count of vertical direction, and T is an error threshold.Yet pixel difference algorithm distinct issues are responsive especially to the motion of noise and camera and object video, cause false-alarm easily.

The basic thought of existing pixel histogramming algorithm (as a kind of scene change detector and the method thereof of China's No. 03147508 patent application announcement) is, substantially remain unchanged same group of continuous scene inner video image overall situation visual information, show that the statistical law on the pixel distribution histogram is, the pixel distribution histogram between the consecutive frame in same group of scene changes will be far smaller than its sudden change at scene handoff boundary place.The process method step of histogram check algorithm generally includes: at first the thresholding of pixel is split and be divided into discrete minizone, and then add up the number of pixels that every width of cloth image falls into each minizone and obtain image pixel distribution histogram on the thresholding space of pixel, differentiate and whether exist scene to switch by detecting the histogrammic mutation statistic of pixel between consecutive frame then.Histogrammic mutation statistic commonly used has point-to-point difference B2B (seeing also formula 3) and histogram intersection to mend CI (seeing also formula 4).

$B2B=\frac{1}{2P}\underset{i}{\mathrm{\Σ}}|H_1\left(i\right)-H_2\left(i\right)|---\left(3\right)$

$\mathrm{CI}=1-\frac{1}{P}\underset{i}{\mathrm{\Σ}}\min (H_1\left(i\right),H_2\left(i\right))---\left(4\right)$

H in the above-mentioned formula ₁And H ₂Represented the histogram of present frame and former frame respectively, P represents H ₁Or H ₂Accumulative total and.With respect to pixel difference check algorithm, pixel histogram method of inspection can reduce false alarm probability.Yet because the method has been abandoned the considering of object video spatial positional information, it can increase miss probability, and the simplest counter-example is to replace in the frame as object video.

Summary of the invention

In order to overcome above-mentioned the deficiencies in the prior art, and provide a kind of switching detection method of video scene, thus solve present switching detection method of video scene cause false declaration, miss probability problem of higher easily.

The technical solution adopted in the present invention is: a kind of switching detection method of video scene is provided, it at first mismatches pixel count to the forward motion compensation piece of current input image and makes statistics with histogram, next motion compensation block that calculates between consecutive frame mismatches pixel count histogram mutation characteristic statistic, carry out the judgement that scene is switched according to described statistic again, and the output court verdict.

Describedly input picture forward motion compensation piece is mismatched pixel count make the step of statistics with histogram and comprise, at first, each present frame is divided into the sub-piece of fixed dimension of non-overlapping copies; Secondly, choose its former frame, calculate the forward motion vector of current sub-block earlier, and calculate the pixel count that mismatches of current sub-block, again this motion compensation block is mismatched pixel count and carry out statistics with histogram according to this forward motion vector as the reference image; Once more, investigate whether to travel through entire image,, then skip to next height piece if not and turn back to and choose the step that reference picture calculates its forward motion vector and rerun if then finish statistics.

The step that the motion compensation block of described calculating current sub-block mismatches pixel count comprises, calculate the respective pixel absolute deviation of current block and reference frame match block earlier, whether differentiate above-mentioned respective pixel absolute deviation more one by one greater than given pixel matching error thresholding, if mismatching the pixel count counter, then current motion compensation block adds 1, otherwise constant.

Described motion compensation block to current sub-block mismatches the step that pixel count carries out statistics with histogram and comprises, check that earlier this mismatches pixel count and falls into location index between its codomain discrete regions, will mismatch the pixel histogram counters to motion compensation block that should interval location index again and add 1.

Described to current sub-block

Motion compensation block mismatch pixel count NMP, pixel absolute deviation PAD and mismatch the computational methods of pixel metering MPC as follows:

$\mathrm{NMP}\left({\stackrel{\mathrm{\ρ}}{X}}_0\right)=\underset{\stackrel{\mathrm{\ρ}}{X}\∈B\left({\stackrel{\mathrm{\ρ}}{X}}_0\right)}{\mathrm{\Σ}}\mathrm{MPC}(f_1\left(\stackrel{\mathrm{\ρ}}{X}\right),f_2(\stackrel{\mathrm{\ρ}}{X}+\stackrel{\mathrm{\ρ}}{D}))$

$\mathrm{PAD}(f_1\left(\stackrel{\mathrm{\ρ}}{X}\right),f_2(\stackrel{\mathrm{\ρ}}{X}+\stackrel{\mathrm{\ρ}}{D}))=|f_1\left(\stackrel{\mathrm{\ρ}}{X}\right)-f_2(\stackrel{\mathrm{\ρ}}{X}+\stackrel{\mathrm{\ρ}}{D})|$

Wherein

Represented visual location of pixels coordinate,

Represented initial pixel to be positioned at

Image block, $\stackrel{\mathrm{\ρ}}{X}\∈B\left({\stackrel{\mathrm{\ρ}}{X}}_0\right)$ Representative The traversal current block Interior all pixels, f ₁And f ₂Represented the pixel value of present frame and former frame respectively, Be the motion vector of image block, T is a pixel matching error thresholding.

The decision method that described scene is switched is if described motion compensation block mismatches pixel histogram mutation characteristic statistic greater than judgement threshold, to think that then present frame is the scene switching point.

Described motion compensation block mismatches pixel histogram mutation characteristic statistic J and calculates gained by point-to-point difference function, and this function is as follows:

$J=\frac{1}{2P}\underset{i}{\mathrm{\Σ}}|H_1\left(i\right)-H_2\left(i\right)|$

H wherein ₁And H ₂Represented the motion compensation block of present frame and former frame to mismatch the pixel count histogram respectively, P represents H ₁Or H ₂Accumulative total and.

Be appreciated that described motion compensation block mismatches pixel count histogram mutation characteristic statistic J and also can calculate gained by the histogram intersection complementary function, this function is as follows:

$J=1-\frac{1}{P}\underset{i}{\mathrm{\Σ}}\min (H_1\left(i\right),H_2\left(i\right))$

H wherein ₁And H ₂Represented the motion compensation block of present frame and former frame to mismatch the pixel count histogram respectively, P represents H ₁Or H ₂Accumulative total and.

Said method also comprised before the motion compensation block between the calculating consecutive frame mismatches pixel histogram mutation characteristic statistic investigates whether present frame is the step of second frame of this group scene, if then skip to next frame and return and restart, then carry out if not and calculate this statistic; In addition, this method also comprises in output whether investigation travels through the step of whole group video sequence after the court verdict, if then end then skips to next frame if not and returns and restart.

Beneficial effect of the present invention is: switching detection method of video scene of the present invention adopts the Differential Detection technology based on motion compensation, by image forward motion compensation piece is mismatched the histogrammic analysis of pixel count, adjudicate the scene switching point, not only can overcome the sensitiveness of consecutive frame coordination pixel difference algorithm to noise and motion, but also can overcome in the conventional pixel intensity statistics histogram analysis decision-making technique ignorance to the object video movable information, reduce the false alarm probability and the miss probability of video scene change detection significantly, promote the discrimination precision and the accuracy of scene detection.

Description of drawings

Fig. 1 is the handling process schematic diagram of switching detection method of video scene of the present invention;

Fig. 2 is that the motion compensation block of switching detection method of video scene of the present invention mismatches pixel statistics with histogram schematic flow sheet.

Embodiment

See also Fig. 1, switching detection method of video scene of the present invention is to remain unchanged substantially same group of continuous scene inner video image overall situation visual information according to principle, show that the regularity of distribution that motion compensation block mismatches on the pixel histogram is, motion compensation block mismatches the histogrammic mutation of pixel and will be far smaller than its sudden change at scene handoff boundary place between the same group of interior consecutive frame of scene.Its concrete enforcement may further comprise the steps:

S101. the forward motion compensation piece of current input image I is mismatched pixel count and make statistics with histogram;

Whether S102. investigate I is second frame of this group scene, if then skip to next frame and return step S101, and execution in step S103 then if not;

S103. the motion compensation block that calculates present frame mismatches the mutation characteristic statistic J of pixel count histogram with respect to its former frame;

S104. carry out the judgement that scene is switched according to J, and the output court verdict: if greater than decision threshold Th (present embodiment Th=0.4), then thinking, J exist scene to switch;

S105. whether investigation travels through whole group video sequence, then skips to next frame if not and return step S101 to re-execute, if then finish.

See also Fig. 2, input picture forward motion compensation piece is mismatched pixel count may further comprise the steps of switching detection method of video scene of the present invention as statistics with histogram:

S201. the fixed dimension that present image I is divided into non-overlapping copies is the sub-piece of M * N (wherein M and N are respectively the pixel count of the horizontal direction and the vertical direction of image block);

S202. choose its former frame as the reference image, calculate the forward motion vector of current sub-block

S203. basis The motion compensation block that calculates current sub-block mismatch pixel count (Number ofMismatching Pixels, NMP);

S204. NMP is carried out statistics with histogram;

S205. whether investigation travels through entire image, then skips to next height piece if not and return step S202 to re-execute, if then finish statistics.

In above-mentioned steps, it is as follows that switching detection method of video scene of the present invention mismatches pixel (NMP sees also formula 5) computational methods to motion compensation block:

At first, calculate current block and reference frame match block respective pixel absolute deviation (Pixel AbsoluteDifference, PAD, see also formula 6), secondly, whether differentiate each PAD one by one greater than given pixel matching error thresholding T (present embodiment T=15), if then mismatch pixel counter (MismatchingPixels Counter, MPC) add 1, otherwise constant (seeing also formula 5 and 6).

$\mathrm{NMP}\left({\stackrel{\mathrm{\ρ}}{X}}_0\right)=\underset{\stackrel{\mathrm{\ρ}}{X}\∈B\left({\stackrel{\mathrm{\ρ}}{X}}_0\right)}{\mathrm{\Σ}}\mathrm{MPC}(f_1\left(\stackrel{\mathrm{\ρ}}{X}\right),f_2(\stackrel{\mathrm{\ρ}}{X}+\stackrel{\mathrm{\ρ}}{D}))---\left(5\right)$

$\mathrm{PAD}(f_1\left(\stackrel{\mathrm{\ρ}}{X}\right),f_2(\stackrel{\mathrm{\ρ}}{X}+\stackrel{\mathrm{\ρ}}{D}))=|f_1\left(\stackrel{\mathrm{\ρ}}{X}\right)-f_2(\stackrel{\mathrm{\ρ}}{X}+\stackrel{\mathrm{\ρ}}{D})|---\left(7\right)$

Wherein

Represented visual location of pixels coordinate,

Represented initial pixel to be positioned at

Image block, $\stackrel{\mathrm{\ρ}}{X}\∈B\left({\stackrel{\mathrm{\ρ}}{X}}_0\right)$ Representative

The traversal current block

Interior all pixels, f ₁And f ₂Represented the pixel value of present frame and former frame respectively,

Be the motion vector of image block, T is a pixel matching error thresholding.

In above-mentioned steps, switching detection method of video scene of the present invention mismatches pixel count (NMP) to motion compensation block, and to carry out the statistics with histogram method as follows:

Check that earlier this NMP falls into I between its codomain [0, M*N] discrete regions _iLocation index i, will mismatch pixel histogram counters H (i) to motion compensation block that should interval location index again and add 1.

In above-mentioned steps, motion compensation block between the described consecutive frame of switching detection method of video scene of the present invention mismatches the calculating of pixel histogram mutation statistic J, can select existing point-to-point difference B2B function (seeing also formula 3) for use, also can select for use existing histogram intersection to mend CI function (seeing also formula 4).

Claims (9)

1. switching detection method of video scene, it is characterized in that: it at first mismatches pixel count to the forward motion compensation piece of current input image and makes statistics with histogram, next motion compensation match block of calculating between consecutive frame mismatches pixel count histogram mutation characteristic statistic, carry out the judgement that scene is switched according to described statistic again, and the output court verdict.

2. switching detection method of video scene as claimed in claim 1, it is characterized in that: describedly current input image forward motion compensation piece is mismatched pixel count make the step of statistics with histogram and comprise, at first each present frame is divided into the sub-piece of fixed dimension of non-overlapping copies; Secondly, choose its former frame, calculate the forward motion vector of current sub-block earlier, and mismatch pixel count, again this motion compensation block is mismatched pixel count and carry out statistics with histogram according to the motion compensation block of this forward motion vector calculating current sub-block as the reference image; Once more, investigate whether to travel through entire image,, then skip to next height piece if not and turn back to and choose the step that reference picture calculates its forward motion vector and rerun if then finish statistics.

3. switching detection method of video scene as claimed in claim 2, it is characterized in that: the step that the motion compensation block of described calculating current sub-block mismatches pixel count comprises, calculate the respective pixel absolute deviation of current sub-block and reference frame match block earlier, whether differentiate above-mentioned respective pixel absolute deviation more one by one greater than given pixel matching error thresholding, if mismatching the pixel count counter, then current motion compensation block adds 1, otherwise constant.

4. switching detection method of video scene as claimed in claim 3, it is characterized in that: describedly current motion compensation block is mismatched the step that pixel count carries out statistics with histogram comprise, check that earlier this motion compensation block mismatches pixel count and falls into location index between its codomain discrete regions, will mismatch the pixel histogram counters to motion compensation block that should interval location index again and add 1.

5. switching detection method of video scene as claimed in claim 4 is characterized in that: described to current sub-block

Motion compensation block mismatch pixel count NMP, pixel absolute deviation PAD and mismatch the computational methods of pixel metering MPC as follows:

$\mathrm{NMP}\left({\stackrel{\mathrm{\ρ}}{X}}_0\right)=\underset{\stackrel{\mathrm{\ρ}}{X}\∈B\left({\stackrel{\mathrm{\ρ}}{X}}_0\right)}{\mathrm{\Σ}}\mathrm{MPC}(f_1\left(\stackrel{\mathrm{\ρ}}{X}\right),f_2(\stackrel{\mathrm{\ρ}}{X}+\stackrel{\mathrm{\ρ}}{D}))$

$\mathrm{PAD}(f_1\left(\stackrel{\mathrm{\ρ}}{X}\right),f_2(\stackrel{\mathrm{\ρ}}{X}+\stackrel{\mathrm{\ρ}}{D}))=|f_1\left(\stackrel{\mathrm{\ρ}}{X}\right)-f_2(\stackrel{\mathrm{\ρ}}{X}+\stackrel{\mathrm{\ρ}}{D})|$

Wherein Represented visual location of pixels coordinate,

Represented initial pixel to be positioned at Image block, $\stackrel{\mathrm{\ρ}}{X}\∈B\left({\stackrel{\mathrm{\ρ}}{X}}_0\right)$ Representative

The traversal current block

Interior all pixels, f ₁And f ₂Represented the pixel value of present frame and former frame respectively,

Be the motion vector of image block, T is a pixel matching error thresholding.

6. as each described switching detection method of video scene of claim 1 to 5, it is characterized in that: the decision method that described scene is switched is, if described motion compensation block mismatches pixel histogram mutation characteristic statistic greater than judgement threshold, think that then present frame is the scene switching point.

7. switching detection method of video scene as claimed in claim 6 is characterized in that: described motion compensation block mismatches pixel histogram mutation characteristic statistic J and calculates gained by point-to-point difference function, and this function is as follows:

$J=\frac{1}{2P}\underset{i}{\mathrm{\Σ}}|H_1\left(i\right)-H_2\left(i\right)|$

H wherein ₁And H ₂Represented the motion compensation block of present frame and former frame to mismatch the pixel count histogram respectively, P represents H ₁Or H ₂Accumulative total and.

8. switching detection method of video scene as claimed in claim 6 is characterized in that: described motion compensation block mismatches pixel histogram mutation characteristic statistic J and calculates gained by the histogram intersection complementary function, and this function is as follows:

$J=1-\frac{1}{P}\underset{i}{\mathrm{\Σ}}\min (H_1\left(i\right),H_2\left(i\right))$

H wherein ₁And H ₂Represented the motion compensation block of present frame and former frame to mismatch the pixel count histogram respectively, P represents H ₁Or H ₂Accumulative total and.

9. switching detection method of video scene as claimed in claim 1, it is characterized in that: this method also comprised before the motion compensation block between the calculating consecutive frame mismatches pixel count histogram mutation characteristic statistic investigates whether present frame is the step of second frame of this group scene, if then skip to next frame and return and restart, then carry out if not and calculate this statistic; In addition, this method also comprises in output whether investigation travels through the step of whole group video sequence after the court verdict, if then end then skips to next frame if not and returns and restart.

Images