CN101409819A - Method for encrypting and deciphering digital camera picture based on voiceprint - Google Patents

Abstract

The invention relates to an encrypting/decrypting method for a digital camera picture based on voice print. The current digital cameras do not have the function of data encrypting generally. The picture encrypting method firstly carries out voice print training on the voice of a user to generate voice print template data; then JPEG coding is carried out on the picture; then the front two characters are modified into sexadecimal data 0xEE and 0xCC; bitwise xor operation is carried out on each byte of the voice print template data with the picture data after coding in sequence. The picture decrypting method firstly carries out matching on ID authentication voice print character and the voice print template; then the picture data of the sexadecimal data 0xEE and 0xCC is corrected to be 0xFFD8; bitwise xor operation is carried out on each byte of the voice print template data with the encrypted picture data in sequence until all the voice print template data is used up. The method is characterized by relative stability and relative uniqueness as well as convenience, safety and hard crack.

Description

A kind of method for encrypting and deciphering digital camera picture based on vocal print

Technical field

The invention belongs to the image processing technique field, be specifically related to a kind of encrypting and decrypting method of the digital camera picture based on the vocal print technology.

Background technology

At present, the digital camera on the market does not generally have data encryption feature, and after the memory device accident on camera or the camera was lost, others can check photo and the video content of having taken on the camera arbitrarily, thereby causes the leakage of individual privacy.

In addition, even adopt identity such as similar bank card password to differentiate that means encrypt camera and captured picture data, fail safe is also not high enough, forge easily and usurp, and use convenient inadequately.

Summary of the invention

The present invention is exactly at the deficiencies in the prior art, provides a kind of vocal print by the people to distinguish people's identity, so to captured picture data encrypt, decryption method.

The present invention includes image ciphering method and picture decryption method.

The concrete steps of image ciphering method are:

(1) sound to the user carries out the vocal print training, and generation vocal print template data also is stored in the data storage.The method of vocal print training is specifically:

1. the speech-sound data sequence to input carries out preliminary treatment, and pretreated concrete grammar is:

A. at first enroll one section k (k 〉=2) speech-sound data sequence in second x (1), x (2) ... x (8000 * k) }, each cumlative energy E constantly of computing voice data, $E=\underset{n=1}{\overset{N}{\mathrm{\Σ}}}{x}^2\left(n\right),$ N=1,2 ... 8000 * k if cumlative energy E, then removes all speech datas before this moment less than default quiet threshold value, keeps the speech-sound data sequence in s (s 〉=1) second after this moment; If remove the not enough s of the speech-sound data sequence second after quiet, then re-enter.

B. utilize formula y (n)=x (n)-α x (n-1) that speech-sound data sequence is carried out preemphasis and handle, wherein α is a pre emphasis factor, 0.9≤α≤1.0, and x is input, y is output.

C. speech-sound data sequence is divided into P frame, each frame comprises 256 speech datas, overlapping 128 speech datas of consecutive frame.

D. the speech data to each frame carries out the windowing smoothing processing.

2. calculate Mel frequency cepstral coefficient (MFCC) vector of pretreated speech-sound data sequence, and it is carried out vector quantization.

The concrete computational process of Mel frequency cepstral coefficient is:

A. the speech data to each frame carries out fast Fourier transform (FFT), obtains linear spectral.

B. calculate the energy spectrum of linear spectral | X| ², and, carry out bandpass filtering to the triangle filter group of this energy spectrum by one group of M Mel yardstick.

C. calculate the logarithm energy S (m) of each filter output, m=0,1 ..., M-1, computing formula is $S\left(m\right)=\ln \left[\underset{k=0}{\overset{M-1}{\mathrm{\Σ}}}{\left|X\right|}^2{H}_m\left(k\right)\right],$ H wherein _m(k) be the frequency response of the triangle filter of m Mel yardstick.

D. the logarithm energy is carried out discrete cosine transform (DCT) and obtain Mel frequency cepstral coefficient vector C (r), r=0,1 ..., M-1, computing formula is $C\left(r\right)=\underset{m=0}{\overset{M-1}{\mathrm{\Σ}}}S\left(m\right)\cos \left[\frac{\mathrm{\πr}(m-0.5)}{M}\right].$

The detailed process of vector quantization is:

Adopt alternative manner to carry out the cluster of Mel frequency cepstral coefficient, every iteration once generates a corresponding characteristic vector.

3. repeat that 1. and 2. L (3≤L≤5) is inferior, the vector center of vector quantization of getting L Mel frequency cepstral coefficient is as the vocal print template data.

(2) with take a picture and carry out JPEG coding;

(3) preceding two bytes of the picture data after will encoding are revised as hexadecimal data 0xEE and 0xCC respectively, as the sign of encrypting photo;

(4) call vocal print template data in the data storage, and with each byte of vocal print template data successively with coding after picture data carry out the step-by-step xor operation since the 3rd byte, till the vocal print template data exhausted, image ciphering was finished thus.

The concrete steps of picture decryption method are:

(1) the authentication voice to user's input carry out the voice preliminary treatment, calculate its Mel frequency cepstral coefficient, the row vector of going forward side by side quantizes, authentication vocal print feature as input, and mate with the vocal print template that is kept in the data storage, the concrete grammar of coupling is: at first call the vocal print template in the data storage, and calculate the Euclidean distance between itself and authentication vocal print feature; If this Euclidean distance, thinks then that voice print matching is successful smaller or equal to preset threshold; If this Euclidean distance, thinks then that vocal print does not match greater than preset threshold; Carry out the deciphering of photo after the voice print matching success.

(2) with preceding two picture datas that byte is hexadecimal data 0xEE and 0xCC, be revised as 0xFFD8, as the beginning flag of jpeg data;

(3) call vocal print template data in the data storage, and with each byte of vocal print template data successively with encrypt after picture data carry out the step-by-step xor operation since the 3rd byte, till the vocal print template data exhausted, picture reverted to common picture.

The inventive method is utilized people's sound characteristic, has relative stability and relative uniqueness, adopts vocal print as the photo encryption technology, have conveniently, safely, the characteristics that are difficult for cracking.The realization of photo encryption function is limited to the increase of camera cost on the camera, and encryption function is very practical.

Description of drawings

Fig. 1 is voice pretreatment process figure;

Fig. 2 is a vocal print feature extraction flow chart;

Fig. 3 is vocal print training flow chart;

Fig. 4 is the voice print matching flow chart;

Fig. 5 is photo encryption flow figure;

Fig. 6 is photo deciphering flow chart.

Embodiment

Method for encrypting and deciphering digital camera picture based on vocal print comprises that voice preliminary treatment, vocal print feature extraction, vocal print features training, voice print matching, photo are encrypted and the photo deciphering.

Referring to Fig. 1, Fig. 2, Fig. 3, Fig. 4, Fig. 5, Fig. 6, describe realization principle of the present invention and workflow in detail.The voice preliminary treatment is to give the vocal print training module after voice are gone quiet, preemphasis and divide the frame operation or the vocal print characteristic extracting module is carried out the vocal print Feature Extraction; The vocal print characteristic extracting module receives pretreated voice, and the Mel yardstick frequency that extracts each frame voice is fallen general coefficient (MFCC), and adopts the iteration vector quantization method to obtain one group of vocal print characteristic vector; The vocal print training then repeatedly calls the voice pretreatment module and the vocal print characteristic extracting module is carried out the vocal print Feature Extraction, and adopts dynamic regular algorithm that the feature of each extraction is trained characteristic vector of regular one-tenth, forms the vocal print template; Voice print matching is to carry out coupling and the authentication of vocal print with the vocal print template of having preserved with voice are carried out after the feature extraction; It is to utilize the vocal print template of keeping that the captured photo behind the JPEG coding is carried out the step-by-step XOR to encrypt that photo is encrypted, and corresponding encryption indicator is set; The photo deciphering is to utilize the vocal print template of keeping that the photo after encrypting is carried out the deciphering of step-by-step XOR, and revises corresponding jpeg file head, the unencryption JPEG photo that formation can directly be browsed.

The inventive method each several part workflow is as described below:

One, voice preliminary treatment

Receive after the voice signal, will carry out preliminary treatment work to voice signal, with convenient follow-up vocal print feature extraction, detailed process as shown in Figure 1.

Specifically, the voice preliminary treatment comprises following detailed process:

A. quiet removal: owing to quiet influence to the vocal print feature is very big, the quiet data that therefore at first needed will import in the voice before carrying out the vocal print feature extraction is removed.In this preferred embodiment, at first for program distribute one section can keep 2 second speech data internal memory, after the user confirms to begin to record, need the voice key of input oneself in 2 seconds.Program is according to the quiet end of quiet threshold decision that presets, and be that 1 second speech data is preserved with the duration of quiet terminal beginning, as the input data of follow-up vocal print feature extraction or training, if the speech data less than of removing after quiet then required the user to re-enter in 1 second.

B. voice preemphasis:, cause the high-frequency energy of voice signal smaller because the effect of vocal cords and lip in the voiced process makes the amplitude at high-frequency resonance peak be lower than the amplitude at low-frequency resonance peak.The purpose of carrying out preemphasis is exactly in order to eliminate the effect of vocal cords and lip, to compensate the HFS of voice signal.The preemphasis formula is:

y(n)＝x(n)-αx(n-1)，0.9≤α≤1.0

Wherein α is a pre emphasis factor, and getting α in this preferred embodiment is 0.97.X is input, and y is output.

C. divide frame: what adopt during voiceprint extraction is the short-time analysis technology, therefore need carry out the branch frame of voice.The length of frame data is got 256 points in this preferred embodiment, and the sample between frame and the frame adopts 1/2 frame overlap mode.

D. windowing: become unintelligible even disappearance for avoiding the gene peak in the cepstrum, need carry out the windowing smoothing processing to voice.Adopt Hamming window in this preferred embodiment.

Through behind the said process, the speech data of input is divided into the windowing short signal sequence of a frame one frame, when carrying out subsequent treatment, takes out speech data frame by frame and carry out respective handling from the data field.

Two, vocal print feature extraction

The vocal print feature extraction is the basis of whole system, and phonetic recognization rate is had very significant effects.The preferred embodiments of the present invention adopt MFCC feature behind vector quantization as the vocal print feature that extracts, detailed process as shown in Figure 2:

A. each frame speech data is carried out obtaining linear spectral after the fast Fourier transform (EFT).

B. ask the energy spectrum of linear spectral, and to the triangle filter group of this energy spectrum by one group of M Mel yardstick, thereby in frequency domain energy spectrum is carried out bandpass filtering.In this preferred embodiment, getting M is 20.

C. calculate the logarithm energy S (m) of each bank of filters output, m=0,1 ..., M-1, computing formula is $S\left(m\right)=\ln \left[\underset{k=0}{\overset{M-1}{\mathrm{\Σ}}}{\left|X\right|}^2{H}_m\left(k\right)\right],$ H wherein _m(k) be the frequency response of the triangle filter of m Mel yardstick.

D. the logarithm energy is carried out discrete cosine transform (DCT) and obtain MFCC characteristic vector C (r), r=0,1 ..., M-1, computing formula is $C\left(r\right)=\underset{m=0}{\overset{M-1}{\mathrm{\Σ}}}S\left(m\right)\cos \left[\frac{\mathrm{\πr}(m-0.5)}{M}\right].$

E. adopt vector quantization method to carry out cluster to all MFCC characteristic vectors that obtain.In this preferred embodiment, adopt alternative manner to carry out the cluster of feature, every iteration once generates a corresponding characteristic vector.In this preferred embodiment, the iterations of employing is 16 times.

Through behind the said process, every frame speech data of input can obtain 20 MFCC features, is 16 * 20 through the characteristic vector dimension that forms after the vector quantization cluster, and wherein each feature needs 4 byte representations, therefore, can generate the vocal print characteristic of 16 * 20 * 4=1280 byte altogether.

Three, vocal print features training

Detailed process as shown in Figure 3, the training of vocal print is repeatedly to call the vocal print characteristic extracting module, forms comparatively stable vocal print feature, and is saved in the data storage.Because the voice that are used for participating in training are fixed and the time is shorter, there are stronger correlation in Application on Voiceprint Recognition and speech text, be that the different terms that same individual says also can be considered to different vocal prints, the user can change the vocal print key at any time according to these characteristics, the fail safe of enhanced system.Certainly, change the vocal print key and need at first delete original key, and training again.

Four, voice print matching

Detailed process adopts the Euclidean distance between vocal print template that fixed threshold relatively stores and new input voice vocal print as shown in Figure 4, and when this distance thinks that voice print matching is successful during less than threshold value, otherwise voice print matching gets nowhere.

Five, photo is encrypted

Because the data storage limited storage space of digital camera,, therefore, in this preferred embodiment, all adopt JPEG compressed format no matter whether all captured photos are encrypted always the user wishes that the many more photos of on same SD card storage are good more.

On the other hand, the user wishes that also the encryption photo kept can't browse after breaking away from camera, thereby reaches the purpose of protection privacy, detailed process as shown in Figure 5:

A) preceding two bytes of revising after encoding of picture data are respectively hexadecimal data 0xEE and 0xCC, as the sign of encrypting photo.

B) by the vocal print template in the file system call data storage, and with each byte of vocal print template data successively with coding after picture data carry out the step-by-step xor operation since the 3rd byte, till the vocal print characteristic of 1280 bytes exhausts in the vocal print template.

C) the photo attribute after will encrypting by file system changes " hiding " into, and this operation makes the general user can't directly see the encryption photo on PC.

Above-mentioned ciphering process has following characteristics: encrypting photo and non-encrypted photo can distinguish easily; Arbitrary photo its data volume before and after encrypting keeps constant fully, and ciphering process can be finished in the photo original position; Be used for carrying out the encrypted secret key long enough, crack and hardly may.

Six, photo deciphering

Photo deciphering is the inverse process that photo is encrypted, therefore, detailed process as shown in Figure 6:

1. at first whether preceding two bytes of interpretation picture data are hexadecimal data 0xEE and 0xCC, are that non-encrypted photo need not deciphering if not then show this photo, if then these two bytes are revised as 0xFFD8, as the beginning flag of jpeg data.

2. the photo attribute after will deciphering by file system changes " non-hiding " into.

3. by the vocal print template in the file system call data storage, and with each byte of vocal print template data successively with encrypt after picture data carry out the step-by-step xor operation since the 3rd byte, till the vocal print characteristic of 1280 bytes exhausts in the vocal print template.

Claims (1)

1, a kind of method for encrypting and deciphering digital camera picture based on vocal print comprises image ciphering method and picture decryption method, it is characterized in that:

The concrete steps of image ciphering method are:

(1) sound to the user carries out the vocal print training, and generation vocal print template data also is stored in the data storage; The method of vocal print training is specifically:

1. the speech-sound data sequence to input carries out preliminary treatment, and pretreated concrete grammar is:

A. at first enroll one period k second speech-sound data sequence x (1), x (2) ... x (8000 * k) }, k 〉=2, each cumlative energy E constantly of computing voice data, $E=\underset{n=1}{\overset{N}{\mathrm{\Σ}}}{x}^2\left(n\right),$ N=1,2 ... 8000 * k if cumlative energy E, then removes all speech datas before this moment less than default quiet threshold value, keeps the speech-sound data sequence in the s second after this moment; If remove the not enough s of the speech-sound data sequence second after quiet, then re-enter s 〉=1;

B. utilize formula y (n)=x (n)-α x (n-1) that speech-sound data sequence is carried out preemphasis and handle, wherein α is a pre emphasis factor, 0.9≤α≤1.0, and x is input, y is output;

C. speech-sound data sequence is divided into P frame, each frame comprises 256 speech datas, overlapping 128 speech datas of consecutive frame;

D. the speech data to each frame carries out the windowing smoothing processing;

2. calculate the Mel frequency cepstral coefficient vector of pretreated speech-sound data sequence, and it is carried out vector quantization; Wherein

The concrete computational process of Mel frequency cepstral coefficient is:

E. the speech data to each frame carries out fast Fourier transform, obtains linear spectral;

F. calculate the energy spectrum of linear spectral | X| ², and, carry out bandpass filtering to the triangle filter group of this energy spectrum by one group of M Mel yardstick;

G. calculate the logarithm energy S (m) of each filter output, m=0,1 ..., M-1, computing formula is $S\left(m\right)=\ln \[\underset{k=0}{\overset{M-1}{\mathrm{\Σ}}}{{\left|X\right|}^{2}H}_m\left(k\right)\],$ H wherein _m(k) be the frequency response of the triangle filter of m Mel yardstick;

H. the logarithm energy is carried out discrete cosine transform (DCT) and obtain Mel frequency cepstral coefficient vector C (r), r=0,1 ..., M-1, computing formula is $C\left(r\right)=\underset{m=0}{\overset{M-1}{\mathrm{\Σ}}}S\left(m\right)\cos \[\frac{\mathrm{\πr}(m-0.5)}{M}\];$

The detailed process of vector quantization is:

Adopt alternative manner to carry out the cluster of Mel frequency cepstral coefficient, every iteration once generates a corresponding characteristic vector;

3. repeat 1. and 2. L time, 3≤L≤5, the vector center of vector quantization of getting L Mel frequency cepstral coefficient is as vocal print mould number of poles certificate;

(2) with take a picture and carry out JPEG coding;

(3) preceding two bytes of the picture data after will encoding are revised as hexadecimal data 0xEE and 0xCC respectively, as the sign of encrypting photo;

(4) call vocal print template data in the data storage, and with each byte of vocal print template data successively with coding after picture data carry out the step-by-step xor operation since the 3rd byte, till the vocal print template data exhausted, image ciphering was finished thus;

The concrete steps of picture decryption method are:

(A) the authentication voice to user's input carry out the voice preliminary treatment, calculate its Mel frequency cepstral coefficient, the row vector of going forward side by side quantizes, authentication vocal print feature as input, and mate with the vocal print template data that is kept in the data storage, the concrete grammar of coupling is: at first call the vocal print template data in the data storage, and calculate the Euclidean distance between itself and authentication vocal print feature; If this Euclidean distance, thinks then that voice print matching is successful smaller or equal to preset threshold; If this Euclidean distance, thinks then that vocal print does not match greater than preset threshold; Carry out the deciphering of photo after the voice print matching success;

(B) with preceding two picture datas that byte is hexadecimal data 0xEE and 0xCC, be revised as 0xFFD8, as the beginning flag of jpeg data;

(C) call vocal print template data in the data storage, and with each byte of vocal print template data successively with encrypt after picture data carry out the step-by-step xor operation since the 3rd byte, till the vocal print template data exhausted, picture reverted to common picture.

Images