Image steganography method based on HSI spatial segmentation algorithm

Abstract

The invention relates to an image steganography method based on an HSI space segmentation algorithm, and the method comprises the steps: obtaining an input color carrier image and to-be-embedded secret information, taking the color carrier image as an original carrier image, decomposing the original carrier image into an HSI color space, segmenting H, S and I components in the HSI color space respectively to obtain segmented carrier images, and carrying out the segmentation of the segmented carrier images, and carrying out edge enhancement on the segmented carrier image to obtain an edge-enhanced carrier image, inputting the edge-enhanced carrier image and the secret information to be embedded into an information hiding network for embedding to obtain a secret-carrying image, and extracting the secret information in the secret-carrying image by using an information extraction network. Therefore, segmentation and edge enhancement processing are independently carried out on each component in the original carrier image in the HSI color space, the carrier image has a better edge enhancement effect, secret information embedding under the condition of conforming to the visual characteristics of human eyes is completed, the special attention of the human eyes on the secret-loaded image is not aroused, and the user experience is improved. And the security of image steganography is facilitated.

Description

Technical Field

[0001] The present invention relates to the field of image steganography, and in particular to an image steganography method based on HSI space segmentation algorithm. Background Art

[0002] As an important sub-field in the new generation of information technology, image steganography is crucial to ensuring information security and achieving privacy protection. In the process of information security protection based on image steganography, secret information needs to be embedded into the carrier image to achieve the hiding of secret information in the carrier image.

[0003] Chinese invention patent CN113538202B discloses an image steganography method and system based on generative steganalysis adversarial. The image steganography method includes constructing a generative steganalysis adversarial model including a generator network, a steganalyzer, a discriminant network and a steganalysis network. The generator network converts an original image into a carrier image. The steganalyzer embeds a secret image into the carrier image to obtain a carrier image. The discriminant network determines the authenticity of the carrier image. The steganalysis network determines whether the carrier image contains secret image information. According to the judgment results of the discriminant network and the steganalysis network, the loss functions of the generator network, the discriminant network and the steganalysis network are respectively constructed. According to the loss functions of the three, the generator network, the discriminant network and the steganalysis network are adversarially trained to obtain a trained generative steganalysis adversarial model. Based on the trained generative steganalysis adversarial model, a steganalysis image corresponding to the original image is obtained to generate a carrier image with high visual quality, thereby improving the security of hiding information of the carrier image.

[0004] However, the image steganography method based on generative steganography adversarial disclosed in the above-mentioned invention patent CN113538202B has shortcomings: only multiple groups of processing units are used in the generative network, but the important characteristics of the image cannot be paid special attention to, which easily leads to the gradient disappearance phenomenon and makes it difficult to conduct adversarial training, which ultimately reduces the security of hiding secret information. Summary of the invention

[0005] The technical problem to be solved by the present invention is to provide an image steganography method based on the HSI space segmentation algorithm to improve the security of hiding secret information in view of the above-mentioned prior art.

[0006] The technical solution adopted by the present invention to solve the above technical problem is: an image steganography method based on the HSI space segmentation algorithm, characterized in that it includes the following steps: Step S1, obtaining an input color carrier image and secret information to be embedded; Step S2, taking the acquired color carrier image as the original carrier image, and decomposing the original carrier image into an HSI color space including an H component, an S component and an I component; Step S3, segmenting the H component, S component and I component in the HSI color space obtained after decomposing the original carrier image, to obtain a carrier image after segmentation; Step S4, performing edge enhancement processing on the carrier image after segmentation processing to obtain a carrier image after edge enhancement processing; Step S5, inputting the edge-enhanced carrier image and the secret information to be embedded into the information hiding network for embedding operation, so as to embed the secret information into the edge-enhanced carrier image to obtain the secret image; Step S6, using the information extraction network to extract secret information from the secret image to obtain the embedded secret information.

[0007] Improved, in the image steganography method based on the HSI space segmentation algorithm, in the step S3, the process of segmenting the H component, the S component and the I component in the HSI color space obtained after the original carrier image is decomposed includes the following steps: Step S31, using the S component to distinguish between a high saturation area and a low saturation area in the HSI color space; Step S32, segmenting the high saturation area using the H component; Step S33, using the I component to segment the low saturation area.

[0008] Further improvement, in the image steganography method based on the HSI space segmentation algorithm, in step S31, the S component is used to adopt the minimum fuzzy divergence threshold within the class to distinguish the high saturation area and the low saturation area; wherein the minimum fuzzy divergence is marked as D(P:Q): ; ; ; , , , ; , , ; ; ; ; ; ; ; Where P is the high saturation area on the S component area, Q is the low saturation area on the S component area, Card(S) is the cardinality of the S component area, s is the image pixel value on the S component area, D(P,Q;s) represents the symmetric cross entropy of the high saturation area P relative to the low saturation area Q, and D(Q,P;s) represents the symmetric cross entropy of the low saturation area Q relative to the high saturation area P. is the membership function of the high saturation region P, is the membership function of Q in the low saturation region; is the non-membership function of the high saturation region P, is the non-membership function of Q in the low saturation region; M P is the average pixel value of the high saturation area P, M Q is the average pixel value of Q in the low saturation area, is the sum of the number of pixels whose pixel values ​​are all s in the high saturation area P, L P Represents the maximum value of the pixel value s in the high saturation area P, is the sum of the number of pixels whose pixel values ​​are all s in the low saturation region Q, L Q Represents the maximum value of the pixel value s in the low saturation area Q.

[0009] Furthermore, in the image steganography method based on the HSI space segmentation algorithm, both step S32 and step S33 use the K-means clustering method to perform their respective corresponding segmentation operations; wherein, the constraints for using the K-means clustering method to perform the corresponding segmentation operation are as follows: ; ; Among them, E P represents the sum of the squares of the distances between all feature points in the high saturation area P and their corresponding cluster means, x P =(x1,x2) is the coordinate of a characteristic point in the high saturation region P, g P (x P ) represents the feature point coordinate x P The eigenvalue at position, It represents the set of feature points assigned to class j after the i-th iteration for the high saturation region P, i ≥ 1; represents the mean value of class j after the i+1th iteration in the high saturation region P; K P represents the total number of clusters in the high saturation region P, and the corresponding k here represents the cluster number in the high saturation region P; E Q represents the sum of the squares of the distances between all feature points in the low saturation region Q and their corresponding cluster means, x Q =(x1,x2) is the coordinate of a characteristic point in the low saturation region Q, g Q (xQ ) represents the feature point coordinate x Q The eigenvalue at position, It represents the set of feature points assigned to class j after the i-th iteration for the low saturation region Q, i ≥ 1; represents the mean value of class j after the i+1th iteration in the low saturation region Q; K Q represents the total number of clusters in the low saturation region Q, and the corresponding k here represents the cluster number in the low saturation region Q.

[0010] Furthermore, in the image steganography method based on the HSI space segmentation algorithm, the process of using the K-means clustering method to perform the corresponding region segmentation operation is as follows: Step a1, based on the elbow method and the error square sum change trend, determine and select the pixel value mean of K initial classes in the area to be segmented; wherein: ; Among them, SSE represents the sum of squared errors, C k represents the kth cluster in the clustering cluster, K is the value corresponding to the turning point of the error square sum SSE change trend, and K is the required number of clusters; m k Represents cluster C k The value of the center pixel, c is the cluster C k The sample data in; the area to be segmented is marked as G, G is a high saturation area P or a low saturation area Q; Step a2: when executing the i-th iteration, assign each feature point to the corresponding cluster in the K initial classes according to the clustering criterion; wherein, for any feature point in the area G to be segmented, ,if , , , , then the feature point x is assigned to the cluster that is located in the K initial clusters and whose mean is closest to the feature point x; Step a3: for all clusters in the K initial classes, update the mean of all numerical points in each cluster; the mean of the cluster j corresponding to the region G to be segmented after the i+1th iteration is marked as : ; Among them, N j is the number of different numerical points in all clusters corresponding to the segmentation area G to be performed; Step a4: for all clusters in the area to be segmented, make a judgment based on all the updated means obtained: When the mean pixel value of all pixels in any cluster satisfies , it indicates that the cluster segmentation operation of the current segmentation operation area has been completed and no iterative operation is performed; otherwise, go to step a2 and continue the next iterative operation.

[0011] Further improvement, in the image steganography method based on the HSI space segmentation algorithm, in the step S4, the process of edge enhancement processing of the carrier image after segmentation processing is as follows: Step b1, performing median filter smoothing on the boundaries of each segmented area in the carrier image after segmentation processing, and obtaining corresponding segmented areas after smoothing processing; Step b2, using a 3×3 filter to process the boundaries of each smoothed segmented region, respectively, to obtain corresponding boundary processed segmented regions; Step b3, re-superimposing each segmented area after boundary processing onto the original carrier image to obtain a carrier image after edge enhancement processing.

[0012] Furthermore, in the image steganography method based on the HSI space segmentation algorithm, in the step S5, the process of obtaining the secret image is as follows: Step c1, pre-constructing an information hiding network; wherein the network structure of the information hiding network is composed of a steganalyzer and a decoder, and the steganalyzer has a convolutional mixed domain attention module; the basic framework of the steganalyzer includes a first group of convolutional blocks, a second group of convolutional blocks, a third group of convolutional blocks, a fourth group of convolutional blocks and a fifth group of convolutional blocks arranged in sequence, the first group of convolutional blocks includes a convolutional layer and a ReLU activation function, and the remaining four groups of convolutional blocks all use a convolutional mixed domain attention module; Step c2, inputting the edge-enhanced carrier image into the steganalysis encoder of the information hiding network, and sequentially passing through five groups of downsampling encodings of the basic framework in the steganalysis encoder to obtain a downsampled encoded image; wherein, the first group of convolution blocks in the five groups of downsampling encodings of the basic framework includes a convolution layer and a ReLU activation function, and the last four groups of convolution blocks all use a convolution mixed domain attention module; Step c3, the obtained down-sampled encoded image is sequentially subjected to five groups of up-sampling encodings of the steganalysis encoder basic framework of the information hiding network to obtain an up-sampled encoded image, and the up-sampled encoded image is used as the secret image; wherein, the first four groups of convolution blocks in the five groups of up-sampling encodings of the basic framework adopt the convolution mixed domain attention module, and the last group of convolution blocks includes a deconvolution layer and a ReLU activation function.

[0013] Further improvement, in the image steganography method based on the HSI space segmentation algorithm, after executing step S6, it also includes: calculating the loss function between the steganographic image and the original carrier image; wherein the loss function is calculated as follows: L = MSE(I,I'); Where I represents the original carrier image, I' represents the encrypted image, and MSE(I, I') represents the mean square error between the original carrier image I and the encrypted image I'.

[0014] Compared with the prior art, the advantages of the present invention are as follows: the image steganography method based on the HSI space segmentation algorithm of the present invention obtains the input color carrier image and the secret information to be embedded, takes the color carrier image as the original carrier image, decomposes the original carrier image into the HSI color space, and respectively segments the H component, S component and I component in the HSI color space to obtain the carrier image after segmentation, performs edge enhancement processing on the carrier image after segmentation to obtain the carrier image after edge enhancement processing, and then inputs the carrier image after edge enhancement processing and the secret information to be embedded into the information hiding network for embedding operation to obtain the secret image, and uses the information extraction network to extract the secret information in the secret image to obtain the embedded secret information. In this way, the components in the original carrier image are independently segmented and edge enhanced in the HSI color space, and the obtained carrier image has a better edge enhancement effect, and the secret information embedding is completed in accordance with the visual characteristics of the human eye, so that the human eye will not pay special attention to the secret image, which is more conducive to the security of image steganography. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 This is a flow chart of an image steganography method based on the HSI space segmentation algorithm in an embodiment of the present invention; Figure 2 is the original carrier image (i.e., the input color image) in the embodiment of the present invention; Figure 3 is the encrypted image after embedding in the embodiment of the present invention; Figure 4 Schematic diagram of the basic framework of the steganographic encoder in an embodiment of the present invention. DETAILED DESCRIPTION

[0016] The present invention is further described in detail below with reference to the accompanying drawings.

[0017] This embodiment provides an image steganography method based on the HSI space segmentation algorithm. Figure 1 As shown, the image steganography method based on the HSI space segmentation algorithm of this embodiment includes the following steps S1 to S6: Step S1, obtaining an input color carrier image and secret information to be embedded; wherein the input color carrier image (that is, the original carrier image) is referred to as Figure 2 As shown, the secret information to be embedded is the text information "NUIST1"; Step S2, taking the acquired color carrier image as the original carrier image, and decomposing the original carrier image into an HSI color space including an H component, an S component, and an I component; wherein the H component is the hue (Hue) in the color space, the S component is the saturation (Saturation or Chroma) in the color space, and the I component is the brightness (Intensity or Brightness) in the color space; In this embodiment, the partial program code segment for decomposing the original carrier image into the HSI color space is as follows: H,S,I = rgb_to_hsi(carrier) combined_label = fuzzy_partition(H,S,I) feature_vectors = np.stack([H.flatten(),S.flatten(),I.flatten()], axis=1); Step S3, segmenting the H component, S component and I component in the HSI color space obtained after the original carrier image is decomposed, to obtain a carrier image after segmentation; wherein, in step S3 of this embodiment, the process of segmenting the H component, S component and I component in the HSI color space of the original carrier image includes the following steps S31-S33: Step S31, using the S component to distinguish a high saturation area and a low saturation area in the HSI color space; wherein the high saturation area on the S component is marked as P, and the low saturation area on the S component is marked as Q; Step S32, segmenting the high saturation area using the H component; Step S33, segmenting the low saturation area using the I component; Step S4, performing edge enhancement processing on the carrier image after segmentation processing to obtain the carrier image after edge enhancement processing; wherein, in this embodiment, the partial program code section corresponding to the carrier image after edge enhancement processing is as follows: import cv2 import numpy as np def smooth_boundaries_and_overlay(label_map, carrier_image): label_map_uint8 = label_map.astype(np.uint8) smoothed = cv2.medianBlur(label_map_uint8, 3) edges = cv2.Canny(smoothed, 100, 200) carrier_with_edges = carrier_image.copy() carrier_with_edges[edges>0]= [255, 0, 0] return smoothed, carrier_with_edges Step S5, inputting the edge-enhanced carrier image and the secret information to be embedded into the information hiding network for embedding operation, so as to embed the secret information into the edge-enhanced carrier image to obtain a secret image; the secret image here refers to Figure 3 As shown; Step S6, using the information extraction network to extract secret information from the secret image to obtain the embedded secret information.

[0018] The main program code segment of the image steganography method based on the HSI space segmentation algorithm in this embodiment is given below: # main.py # -*- coding: utf-8 -*- import glob import torch import numpy as np from torch.utils.data import DataLoader from PIL import Image from dataset import StegoDataset from network import UNetWithMixedDomainAttention from train import train_model from inference import embed_secret, extract_secret if __name__ == '__main__': device = 'cuda' if torch.cuda.is_available() else 'cpu' # Get data path image_paths = sorted(glob.glob('images / *.png')) message_paths = sorted(glob.glob('messages / *.txt')) # Check if enough data has been loaded if not image_paths: raise FileNotFoundError("No images found in the 'images / 'directory.") if not message_paths: raise FileNotFoundError("No messages found in the 'messages / 'directory.") print(f"Found {len(image_paths)} images and {len(message_paths)}messages.") # Create dataset and DataLoader dataset=StegoDataset(image_paths, message_paths, text_img_size=(32,32), max_txt_len=1024) dataloader = DataLoader(dataset, batch_size=1, shuffle=True) # Define and train the model model = UNetWithMixedDomainAttention(n_channels=4, secret_classes=256) print("Training the model...") model = train_model(model, dataloader, epochs=10, lr=1e-3, device=device) print("Training completed.") # Embedding phase print("Embedding secret message...") test_carrier = 'images / image1.png' # Test carrier image test_message = 'messages / message1.txt' # Test text message # Embed secret information and get the image of the superimposed boundary stego_img, encoded_length, carrier_with_edges = embed_secret(model, test_carrier,test_message, text_img_size=(32, 32), device=device) Image.fromarray(stego_img).save('stego_output.png') Image.fromarray((carrier_with_edges*255).astype(np.uint8)).save('carrier_with_edges.png') print(f"Secret embedded. Encoded length: {encoded_length}") print("Stego image saved to 'stego_output.png'.") print("Carrier with edges saved to 'carrier_with_edges.png'.") # Save the actual embedding length to a file (for use in the extraction phase) with open('encoded_length.txt', 'w') as f: f.write(str(encoded_length)) # Extraction phase print("Extracting secret message...") with open('encoded_length.txt', 'r') as f: encoded_length = int(f.read().strip()) # load embedded length extracted_text, smoothed_labels, extracted_with_edges = extract_secret(model, 'stego_output.png', text_img_size=(32, 32), encoded_length=encoded_length, device=device) # Save the smoothed label image and boundary overlay image Image.fromarray((smoothed_labels*255).astype(np.uint8)).save('smoothed_labels.png') Image.fromarray((extracted_with_edges*255).astype(np.uint8)).save('extracted_with_edges.png') print("Smoothed labels saved to 'smoothed_labels.png'.") print("Extracted image with edges saved to 'extracted_with_edges.png'.") # Save the extracted information with open('extracted_secret.txt', 'w', encoding='utf-8') as f: f.write(extracted_text) print("Extracted text saved to 'extracted_secret.txt'.") # Print the extraction results and compare them with the original message with open(test_message, 'r', encoding='utf-8') as f: original_message = f.read() print(f"Original text: {original_message.strip()}") print(f"Extracted text: {extracted_text.strip()}") if original_message.strip() == extracted_text.strip(): print("Success: Extracted text matches the original message.") else: print("Error: Extracted text does not match the original message.") Specifically, in the above step S31 of this embodiment, the S component is used to distinguish the high saturation area from the low saturation area by adopting the minimum fuzzy divergence threshold within the class; wherein the minimum fuzzy divergence here is marked as D(P:Q): ; ; ; , , , ; , , ; ; ; ; ; ; ; Where P is the high saturation area on the S component area, Q is the low saturation area on the S component area, Card(S) is the cardinality of the S component area, s is the image pixel value on the S component area, D(P,Q;s) represents the symmetric cross entropy of the high saturation area P relative to the low saturation area Q, and D(Q,P;s) represents the symmetric cross entropy of the low saturation area Q relative to the high saturation area P. is the membership function of the high saturation region P, is the membership function of Q in the low saturation region; is the non-membership function of the high saturation region P, is the non-membership function of Q in the low saturation region; M P is the average pixel value of the high saturation area P, M Q is the average pixel value of Q in the low saturation area, is the sum of the number of pixels whose pixel values ​​are all s in the high saturation area P, L P Represents the maximum value of the pixel value s in the high saturation area P, is the sum of the number of pixels whose pixel values ​​are all s in the low saturation region Q, L Q Represents the maximum value of the pixel value s in the low saturation area Q, Indicates the calculation results Find the minimum value.

[0019] Specifically, in the above step S32 and step S33 of this embodiment, both step S32 and step S33 use the K-means clustering method to perform their respective corresponding segmentation operations. The constraints for using the K-means clustering method to perform the corresponding segmentation operation are as follows: ; ; Among them, E P represents the sum of the squares of the distances between all feature points in the high saturation area P and their corresponding cluster means, x P =(x1,x2) is the coordinate of a characteristic point in the high saturation region P, g P (x P ) represents the feature point coordinate x P The eigenvalue at position, It represents the set of feature points assigned to class j after the i-th iteration for the high saturation region P, i ≥ 1; represents the mean value of class j after the i+1th iteration in the high saturation region P; K P represents the total number of clusters in the high saturation region P; ; E Q represents the sum of the squares of the distances between all feature points in the low saturation region Q and their corresponding cluster means, x Q =(x1,x2) is the coordinate of a characteristic point in the low saturation region Q, g Q (x Q ) represents the feature point coordinate x Q The eigenvalue at position, It represents the set of feature points assigned to class j after the i-th iteration for the low saturation region Q, i ≥ 1; represents the mean value of class j after the i+1th iteration in the low saturation region Q; K Q represents the total number of clusters in the low saturation region Q; ; It should be noted that the process of using the K-means clustering method to perform the corresponding region segmentation operation is as follows: steps a1 to a4: Step a1, based on the elbow method and the error square sum change trend, determine and select the pixel value mean of K initial classes in the area to be segmented; wherein: ; Among them, SSE represents the sum of squared errors, C k represents the kth cluster in the clustering cluster, K is the value corresponding to the turning point of the error square sum SSE change trend, and K is the required number of clusters; m k Represents cluster C k The value of the center pixel, c is the cluster C k The sample data in; the area to be segmented is marked as G, G is a high saturation area P or a low saturation area Q; For example, when the K-means clustering method is needed to perform a segmentation operation on a high saturation region P, the segmentation operation region to be performed here is G=P; correspondingly, when the K-means clustering method is needed to perform a segmentation operation on a low saturation region Q, the segmentation operation region to be performed here is G=Q; In this embodiment, the program code section for calculating the initial number of classes K is as follows: # Use the elbow method to dynamically determine the best K value optimal_k, _ = elbow_method(feature_vectors, max_k=10) initial_labels,_=k_means_refine(feature_vectors, combined_label.flatten(),K=optimal_k) refined_labels = initial_labels.reshape(H.shape) Step a2: when executing the i-th iteration, assign each feature point to the corresponding cluster in the K initial classes according to the clustering criterion; wherein, for any feature point in the area G to be segmented, ,if , , , , then the feature point x is assigned to the cluster that is located in the K initial clusters and whose mean is closest to the feature point x; Step a3: for all clusters in the K initial classes, update the mean of all numerical points in each cluster; the mean of the cluster j corresponding to the region G to be segmented after the i+1th iteration is marked as : ; Among them, N j is the number of different numerical points in all clusters corresponding to the segmentation area G to be performed; Step a4: for all clusters in the area to be segmented, make a judgment based on all the updated means obtained: When the mean pixel value of all pixels in any cluster satisfies , it indicates that the cluster segmentation operation of the current segmentation operation area has been completed and no iterative operation will be performed; otherwise, go to step a2 and continue the next iterative operation.

[0020] In addition, in step S4 of this embodiment, the process of performing edge enhancement processing on the carrier image after segmentation processing is as follows: steps b1 to b3: Step b1, performing median filter smoothing on the boundaries of each segmented area in the carrier image after segmentation processing, and obtaining corresponding segmented areas after smoothing processing; Step b2, using a 3×3 filter to process the boundaries of each smoothed segmented region, respectively, to obtain corresponding boundary processed segmented regions; Step b3, re-superimposing each segmented area after boundary processing onto the original carrier image to obtain a carrier image after edge enhancement processing.

[0021] In step S5 of this embodiment, the process of obtaining the encrypted image is as follows: steps c1 to c3: Step c1, pre-constructing an information hiding network; wherein the network structure of the information hiding network is composed of a steganalyzer and a decoder, and the steganalyzer has a convolutional mixed domain attention module; the basic framework of the steganalyzer includes a first group of convolutional blocks conv1, a second group of convolutional blocks conv2, a third group of convolutional blocks conv3, a fourth group of convolutional blocks conv4 and a fifth group of convolutional blocks conv5 which are arranged in sequence, the first group of convolutional blocks conv1 includes a convolutional layer and a ReLU activation function, and the remaining four groups of convolutional blocks (i.e., the second group of convolutional blocks conv2 to the fifth group of convolutional blocks conv5) all adopt a convolutional mixed domain attention module; wherein, the basic framework of the steganalyzer in this embodiment, see Figure 4 As shown; Step c2, input the carrier image after edge enhancement processing into the steganalysis encoder of the information hiding network, and sequentially pass through five groups of downsampling encodings of the basic framework in the steganalysis encoder to obtain a downsampled encoded image; wherein, the first group of convolution blocks (i.e., the first group of convolution blocks conv1) in the process of passing through the five groups of downsampling encodings of the basic framework includes a convolution layer and a ReLU activation function, and the last four groups of convolution blocks (i.e., the second group of convolution blocks conv2 to the fifth group of convolution blocks conv5) all use a convolution mixed domain attention module; Step c3, the obtained down-sampled encoded image is sequentially subjected to five groups of up-sampling encodings of the steganalysis encoder basic framework of the information hiding network to obtain an up-sampled encoded image, and the up-sampled encoded image is used as the secret image; wherein, the first four groups of convolution blocks in the five groups of up-sampling encoding processes of the basic framework adopt convolutional mixed domain attention modules (i.e., the second group of convolution blocks conv2 to the fifth group of convolution blocks conv5), and the last group of convolution blocks (i.e., the first group of convolution blocks conv1) includes a deconvolution layer and a ReLU activation function.

[0022] Of course, in the image steganography method based on the HSI space segmentation algorithm of this embodiment, according to actual needs, the image steganography method can also further include: calculating the loss function between the secret image and the original carrier image after executing step S6; wherein the loss function is calculated as follows: L = MSE(I,I'); Where I represents the original carrier image, I' represents the encrypted image, and MSE(I, I') represents the mean square error between the original carrier image I and the encrypted image I'.

[0023] In the image steganography method based on the HSI space segmentation algorithm of this embodiment, by obtaining the input color carrier image and the secret information to be embedded, the color carrier image is used as the original carrier image and decomposed into the HSI color space, the H, S and I components in the HSI color space are segmented to obtain the carrier image after segmentation processing, the edge enhancement is performed on the carrier image after segmentation processing to obtain the carrier image after edge enhancement processing, the carrier image after edge enhancement processing and the secret information to be embedded are input into the information hiding network for embedding to obtain the secret image, and the secret information in the secret image is extracted by using the information extraction network. In this way, each component in the original carrier image is segmented and edge enhanced independently in the HSI color space, and the obtained carrier image has a better edge enhancement effect, and the secret information is embedded in accordance with the visual characteristics of the human eye, so that the human eye will not pay special attention to the secret image, which is more conducive to the security of image steganography.

[0024] Although the preferred embodiments of the present invention are described in detail above, it should be clearly understood that various modifications and variations are possible for those skilled in the art. Any modification, equivalent substitution, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. The image steganography method based on HSI space segmentation algorithm is characterized by: The following steps are involved: Step S1, obtaining an input color carrier image and secret information to be embedded; Step S2, taking the acquired color carrier image as the original carrier image, and decomposing the original carrier image into an HSI color space including an H component, an S component and an I component; Step S3, segmenting the H component, S component and I component in the HSI color space obtained after decomposing the original carrier image, to obtain a carrier image after segmentation; Step S4, performing edge enhancement processing on the carrier image after segmentation processing to obtain a carrier image after edge enhancement processing; Step S5, inputting the edge-enhanced carrier image and the secret information to be embedded into the information hiding network for embedding operation, so as to embed the secret information into the edge-enhanced carrier image to obtain the secret image; Step S6, using the information extraction network to extract secret information from the secret image to obtain the embedded secret information.

2. The image steganography method based on the HSI space segmentation algorithm according to claim 1 is characterized in that: In step S3, the process of segmenting the H component, S component and I component in the HSI color space obtained after decomposing the original carrier image includes the following steps: Step S31, using the S component to distinguish between a high saturation area and a low saturation area in the HSI color space; Step S32, segmenting the high saturation area using the H component; Step S33, using the I component to segment the low saturation area.

3. The image steganography method based on the HSI space segmentation algorithm according to claim 2 is characterized in that: In step S31, the S component is used to distinguish the high saturation area from the low saturation area by adopting the minimum fuzzy divergence threshold within the class; wherein the minimum fuzzy divergence is marked as D(P:Q): ； ； ； ， ， ， ； ， ， ； ； ； ； ； ； ； Where P is the high saturation area on the S component area, Q is the low saturation area on the S component area, Card(S) is the cardinality of the S component area, s is the image pixel value on the S component area, D(P,Q;s) represents the symmetric cross entropy of the high saturation area P relative to the low saturation area Q, and D(Q,P;s) represents the symmetric cross entropy of the low saturation area Q relative to the high saturation area P. is the membership function of the high saturation region P, is the membership function of Q in the low saturation region; is the non-membership function of the high saturation region P, is the non-membership function of Q in the low saturation region; M P is the average pixel value of the high saturation area P, M Q is the average pixel value of Q in the low saturation area, is the sum of the number of pixels whose pixel values ​​are all s in the high saturation area P, L P Represents the maximum value of the pixel value s in the high saturation area P, is the sum of the number of pixels whose pixel values ​​are all s in the low saturation region Q, L Q Represents the maximum value of the pixel value s in the low saturation area Q.

4. The image steganography method based on the HSI space segmentation algorithm according to claim 3 is characterized in that: The step S32 and the step S33 both use the K-means clustering method to perform their respective corresponding segmentation operations; wherein, the constraints for using the K-means clustering method to perform the corresponding segmentation operations are as follows: ； ； Among them, E P represents the sum of the squares of the distances between all feature points in the high saturation area P and their corresponding cluster means, x P =(x1,x2) is the coordinate of a characteristic point in the high saturation region P, g P (x P ) represents the feature point coordinate x P The eigenvalue at position, It represents the set of feature points assigned to class j after the i-th iteration for the high saturation region P, i ≥ 1; represents the mean value of class j after the i+1th iteration in the high saturation region P; K P represents the total number of clusters in the high saturation region P; E Q represents the sum of the squares of the distances between all feature points in the low saturation region Q and their corresponding cluster means, x Q =(x1,x2) is the coordinate of a characteristic point in the low saturation region Q, g Q (x Q ) represents the feature point coordinate x Q The eigenvalue at position, It represents the set of feature points assigned to class j after the i-th iteration for the low saturation region Q, i ≥ 1; represents the mean value of class j after the i+1th iteration in the low saturation region Q; K Q Represents the total number of clusters in the low saturation region Q.

5. The image steganography method based on the HSI space segmentation algorithm according to claim 4 is characterized in that: In step S4, the process of performing edge enhancement processing on the carrier image after segmentation processing is as follows: Step b1, performing median filter smoothing on the boundaries of each segmented area in the carrier image after segmentation processing, and obtaining corresponding segmented areas after smoothing processing; Step b2, using a 3×3 filter to process the boundaries of each smoothed segmented region, respectively, to obtain corresponding boundary processed segmented regions; Step b3, re-superimposing each segmented area after boundary processing onto the original carrier image to obtain a carrier image after edge enhancement processing.

6. The image steganography method based on the HSI space segmentation algorithm according to claim 5 is characterized in that: In step S5, the process of obtaining the encrypted image is as follows: Step c1, pre-constructing an information hiding network; wherein the network structure of the information hiding network is composed of a steganalyzer and a decoder, and the steganalyzer has a convolutional mixed domain attention module; the basic framework of the steganalyzer includes a first group of convolutional blocks, a second group of convolutional blocks, a third group of convolutional blocks, a fourth group of convolutional blocks and a fifth group of convolutional blocks arranged in sequence, the first group of convolutional blocks includes a convolutional layer and a ReLU activation function, and the remaining four groups of convolutional blocks all use a convolutional mixed domain attention module; Step c2, inputting the edge-enhanced carrier image into the steganalysis encoder of the information hiding network, and sequentially passing through five groups of downsampling encodings of the basic framework in the steganalysis encoder to obtain a downsampled encoded image; wherein, the first group of convolution blocks in the five groups of downsampling encodings of the basic framework includes a convolution layer and a ReLU activation function, and the last four groups of convolution blocks all use a convolution mixed domain attention module; Step c3, the obtained down-sampled encoded image is sequentially subjected to five groups of up-sampling encodings of the steganalysis encoder basic framework of the information hiding network to obtain an up-sampled encoded image, and the up-sampled encoded image is used as the secret image; wherein, the first four groups of convolution blocks in the five groups of up-sampling encodings of the basic framework adopt the convolution mixed domain attention module, and the last group of convolution blocks includes a deconvolution layer and a ReLU activation function.

7. The image steganography method based on the HSI space segmentation algorithm according to claim 6 is characterized in that: After executing step S6, the method further includes: calculating a loss function between the encrypted image and the original carrier image; wherein the loss function is calculated as follows: L = MSE(I,I'); Where I represents the original carrier image, I' represents the encrypted image, and MSE(I, I') represents the mean square error between the original carrier image I and the encrypted image I'.

Images