CN105825476A - Rapid image scanning and mosaicing mode applied to DSP - Google Patents

Abstract

The invention discloses a rapid image scanning and mosaicing mode applied to a DSP. The rapid image scanning and mosaicing mode comprises the following steps that 1, a to-be-mosaiced image set is acquired, and a first image and a second image are selected to be mosaiced, wherein a datum line and a matching range are firstly determined; 2, a difference of every two adjacent pixels on the datum line is calculated; 3, an optimal matching position is determined on a matching line, wherein the matching line is located within the matching range and has the same height as the datum line; 4, mosaicing is achieved, wherein the portion, on the right side of the datum line, of the first image is replaced with the portion, on the right side of the matching line, of the second image after the optimal matching position is found, and then a mosaiced image is obtained; 5, the mosaiced image obtained in the step 4 serves as the first image, a new image is read to serve as the second image, the step 1-the step 4 are repeatedly conducted, and then mosaicing on multiple images can be achieved.

Description

A kind of image quick Overlap-scanning mode being applied to DSP

Technical field

The present invention relates to technical field of image processing, the image being specifically related to a kind of DSP of being applied to quickly scans Connecting method.

Background technology

At present, image mosaic is typically necessary and carries out images match.The matching process of image typically has based on ash The registration of degree, the registration etc. of feature based.Existing frequently-used mode is all the registration mode using feature based. In the technical scheme of existing feature based registration mode, generally use Harris Corner Detection, SIFT yardstick Invariant features conversion scheduling algorithm realizes image mosaic.Wherein Harris Corner Detection, its principle is use one The feature of individual second-order matrix detects characteristic point, and its feature extracted has rotational invariance, and shortcoming is not Possesses scale invariability；SIFT algorithm is to detect some by Gauss difference space to check that some extreme value positions Characteristic point.SIFT algorithm improves quality and the accuracy of images match, and application also ratio is wide.

But existing imaging effect preferable image mosaic mode needs to realize on high-performance PC, needs height The processing speed of effect and the biggest memory headroom, but the embedded processing systems of similar dsp processor, Suffer from the restriction of processing speed and memory headroom, use conventional image mosaic mode processing speed with And splicing effect aspect is undesirable.

Summary of the invention

Therefore, for above-mentioned problem, the present invention, based on the above-mentioned requirement to dsp system, proposes a kind of fast The panoramic mosaic mode of speed scanning, improves existing joining method, reduces processing speed and internal memory is empty Between restriction.

In order to solve above-mentioned technical problem, thinking of the present invention is as follows: move between two pictures Distance excessive time, then the deformation that can not be ignored between image, may also can with the change of light and shade, Situation especially for customer-centric rotary taking panorama photograph to the right.Now to expect preferably spelling Access node fruit, not only to find out the match point of two pictures, also to carry out picture at deformation rectification and light and shade Reason.However, it is contemplated that when displacement between two pictures is minimum (the most only moving string pixel), then The shooting time interval of two pictures is extremely short, and deformation and light and shade difference between image are the most negligible, just Can be to use the image mosaic mode of a kind of quick scanning.

Concrete, the technical solution adopted in the present invention is, the image of a kind of DSP of being applied to quickly scans spelling Connect mode, comprise the steps:

Step 1: obtain image collection to be spliced, chooses first image and second image splices: First datum line and matching range are determined: first image choosing in image collection and second image, first Looking for a datum line in first image, datum line respectively vacates 10 row pixels from the up-and-down boundary of image, high 460 pixels, from the right margin 320 row pixel of image；Then keep left second image center and determine one Joining scope, matching range size is wide 30 row pixels, high 470 pixels, matching range from image upper below Boundary respectively vacates 5 row pixels；

Step 2: calculate the difference of neighbor on datum line: in matching range by matched line one by one with datum line Mating, the height as wherein matched line has with datum line, matching process is: first in datum line Obtain front 50 pairs of pixels that adjacent two pixel values are maximum, and record the position at this 50 couple value place:

A₁=a₁-a₂；

A₂=a₂-a₃；

……

A₄₅₉=a₄₅₉-a₄₆₀；

The wherein pixel value of every on line on the basis of a [i], A [i] is its difference of adjacent 2,1 i 459.A [i] is carried out sequence from big to small and records front 50 values A [0]～the coordinate bit of A [49] correspondence Put；

Step 3: determine best match position on a match line: ask on then corresponding in matched line position Go out the difference of pixel value two-by-two, and 50 pairs of differences of these 50 pairs of differences Yu datum line are subtracted each other two-by-two take absolutely To value sum, each matched line in matching range repeat above operation, then absolute value sum is minimum Matched line is best match position to be looked for.I.e. ask:

$min\left(\underset{i=1}{\overset{50}{\Σ}}\right|A_i-B_i\left|\right)$

Wherein A_iOn the basis of pixel value in line, B_iFor the pixel value in matched line.Obtain minimum The position of the i corresponding to value, is best match position；

Step 4: realize splicing: find second image after best match position in matched line right part It is substituted on the right side of first graphic based line, obtains splicing result images；

Step 5: the splicing result images next step 4 obtained, as first image, then reads one New images proceeds above step as second image and can splice multiple images to obtain complete Jing Tu；

Step 6: cutting up-and-down boundary: the panoramic picture now obtained good effect, yet with In camera translation motion, upper lower deviation is inevitable, and therefore result images can leave blank on up-and-down boundary, Need to cut out upper and lower blank parts, obtain final result image.

The said method of the application present invention carries out panoramic mosaic to image, and its speed is fast, in user's handheld device When slowly rotating, a new images can be read every a few tens of milliseconds and splice, the most adjacent twi-read Image between difference the least just as above-mentioned example, so splicing effect and reality Spend the highest.To sum up, the present invention uses said method, and it enormously simplify Processing Algorithm, thus greatly reduces Processing speed and the restriction of memory headroom to dsp processor, disclosure is particularly well suited to hand-held embedded set The above-mentioned quick scan mode of standby employing carries out image mosaic, has significantly expanded its application scenarios, has well Practicality.

Accompanying drawing explanation

In conjunction with detailed description of the invention, the present invention is further described.

During image mosaic, when between two pictures, the distance of movement is excessive, then can occur between image The deformation that can not be ignored, may also can be with the change of light and shade, especially for customer-centric to dextrorotation Turn the situation of shooting panoramic photo.Now to expect and preferably splice result, not only to find out two pictures Match point, also to carry out deformation rectification and Lightness disposal to picture.However, it is contemplated that between two pictures (the most only move string pixel) when displacement is minimum, then the shooting time interval of two pictures is extremely short, Deformation and light and shade difference between image are the most negligible (dimension of picture of this project is fixed as 640*480), Such as two pictures displacements only have 10 row pixels, it is possible to use the image mosaic side of a kind of quick scanning Formula.The connecting method of the present invention is applied to the embedded device with DSP as hardware carrier, this embedded device The image acquisition of 30 frames per second can be carried out, therefore in splicing, move in pixel per second actual carrying out The image mosaic of quick scanning set forth in the present invention can be realized in the case of about 300 row.

Concrete, a kind of image quick Overlap-scanning mode being applied to DSP of the present invention, implement step Rapid as follows:

Step 1: first obtain image collection to be spliced, it is then determined that datum line matching range: choose image Two images in set, first look for a datum line in first image, datum line from image upper below Boundary respectively vacates 10 row pixels, high 460 pixels, why vacates 10 row pixels and allows for adjacent two figures Situation about shifting up and down is had during sheet splicing, from the right margin 320 row pixel of image, the most placed in the middle at image It is minimum that position looks for datum line to be because the produced deformation in image center during shooting；Then at second image just In keep left and determine a matching range, range size is wide 30 row pixels, high 470 pixels, matching range from Image up-and-down boundary respectively vacates 5 row pixels, the most just can correct situation about shifting up and down to realize spelling Connect；

Step 2: calculate the difference of neighbor on datum line: the core concept of algorithm is general in matching range Distribution mates with datum line one by one, the height as wherein matched line has with datum line, matching process It is: in datum line, first obtain front 50 pairs of pixels that adjacent two pixel values are maximum, and record this 50 couple The position at value place:

A₁=a₁-a₂；

A₂=a₂-a₃；

……

A₄₅₉=a₄₅₉-a₄₆₀；

The wherein pixel value of every on line on the basis of a [i], A [i] is its difference of adjacent 2,1 i 459.A [i] is carried out sequence from big to small and records front 50 values A [0]～the coordinate bit of A [49] correspondence Put；

Step 3: determine best match position on a match line: obtain two on position corresponding in matched line The difference of two pixel values, and 50 pairs of differences of these 50 pairs of differences with datum line are subtracted each other two-by-two take absolute value Sum, repeats above operation, the then coupling that absolute value sum is minimum to each matched line in matching range Line is best match position to be looked for.I.e. ask:

$min\left(\underset{i=1}{\overset{50}{\Σ}}\right|A_i-B_i\left|\right)$

Wherein A_iOn the basis of pixel value in line, B_iFor the pixel value in matched line.Obtain minimum The position of the i corresponding to value, is best match position；

Step 4: realize splicing: find second image after best match position in matched line right part It is substituted on the right side of first graphic based line, obtains splicing result images；

Step 5: next using result images as first image, then reads a new images as second Image proceeds above step and can splice to obtain panorama sketch to multiple images；

Step 6: cutting up-and-down boundary: the panoramic picture now obtained good effect, yet with In camera translation motion, upper lower deviation is inevitable, and therefore result images can leave blank on up-and-down boundary, Need to cut out upper and lower blank parts, obtain final result image.

In this way image should be carried out panoramic mosaic, its speed is fast, slowly rotates in user's handheld device Time, can every a few tens of milliseconds read a new images splice, the image of the most adjacent twi-read it Between difference the least just as above-mentioned example, so splicing effect the highest with reality degree.

Although specifically show and describe the present invention in conjunction with preferred embodiment, but those skilled in the art Should be understood that in the spirit and scope of the present invention limited without departing from appended claims, in form Above and the present invention can be made a variety of changes in details, be protection scope of the present invention.

Claims (5)

1. the image quick Overlap-scanning mode being applied to DSP, it is characterised in that: comprise the steps:

Step 1: obtain image collection to be spliced, chooses first image and second image splices: First datum line and matching range are determined；

Step 2: calculate the difference of neighbor on datum line；

Step 3: determine best match position on a match line, in wherein matched line is in matching range, and with Datum line has identical height；

Step 4: realize splicing: find second image portion on the right side of matched line after best match position Divide the right side being substituted into first graphic based line, i.e. obtain splicing result images；

Step 5: the splicing result images next step 4 obtained, as first image, then reads one Open new images and repeat step 1-step 4 as second image, can realize multiple images are spliced.

Image the most according to claim 1 quick Overlap-scanning mode, it is characterised in that: also include step 6: The up-and-down boundary of the final splicing result images obtained is carried out cutting.

Image the most according to claim 1 and 2 quick Overlap-scanning mode, it is characterised in that: described step 1 specifically includes following content: first image choosing in image collection and second image, first the Looking for a datum line in one image, datum line respectively vacates 10 row pixels from the up-and-down boundary of image, high 460 pixels, from the right margin 320 row pixel of image；Then keep left second image center and determine one Individual matching range, matching range size is wide 30 row pixels, and high 470 pixels, matching range is from image Up-and-down boundary respectively vacate 5 row pixels.

Image the most according to claim 1 and 2 quick Overlap-scanning mode, it is characterised in that: described step 1 specifically includes following content: mated with datum line one by one by matched line in matching range, wherein Matched line have with datum line as height, matching process is: first obtain adjacent two in datum line Front 50 pairs of pixels that pixel value is maximum, and record the position at this 50 couple value place:

A₁=a₁-a₂；

A₂=a₂-a₃；

……

A₄₅₉=a₄₅₉-a₄₆₀；

The wherein pixel value of every on line on the basis of a [i], A [i] is its difference of adjacent 2,1 i 459；

A [i] is carried out sequence from big to small, and records front 50 values A [0]～the coordinate bit of A [49] correspondence Put.

Image the most according to claim 1 and 2 quick Overlap-scanning mode, it is characterised in that: described step 1 specifically includes following content: obtain the difference of pixel value two-by-two on then corresponding in matched line position, And 50 pairs of differences of these 50 pairs of differences Yu datum line are subtracted each other the sum that takes absolute value, two-by-two to coupling In the range of each matched line repeat above operation, then the matched line that absolute value sum is minimum is and to look for Best match position；I.e. ask:

$min\left(\underset{i=1}{\overset{50}{\mathrm{\Σ}}}\right|A_i-B_i\left|\right);$

Wherein A_iOn the basis of pixel value in line, B_iFor the pixel value in matched line；Obtain minima The position of corresponding i, is best match position.