A feature weight determination method and system for feature fusion image retrieval

Abstract

The application relates to the technical field of image retrieval, and provides a feature weight determination method and system for feature fusion image retrieval. The feature weight determination method for feature fusion image retrieval comprises the following steps: acquiring single feature metrics of each image, constructing a single feature metric set, and performing normalization processing to obtain normalized single feature metrics and the single feature metric set; calculating a discrete degree and uncertainty based on the normalized single feature metrics; calculating a single feature metric evaluation value based on the discrete degree and the uncertainty, so as to construct an evaluation set of all single feature metric evaluation values; calculating a correlation matrix based on the correlation between elements in the normalized single feature metric set; distinguishing redundant metrics and applicable metrics based on the evaluation value and the correlation matrix; and finally, calculating the weight of the redundant metrics and the weight of the applicable metrics to obtain the final weight. The accuracy of feature fusion image retrieval is improved.

Description

Technical Field

[0001] The present invention relates to the field of image retrieval technology, and in particular to a method and system for determining feature weights for feature fusion image retrieval. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] In the field of content-based image retrieval (CBIR), the allocation of feature weights directly affects retrieval accuracy and efficiency. Traditional methods mostly use global fixed weight strategies and adaptive weight strategies.

[0004] Among them, the global fixed weight strategy uniformly assigns preset weights to all query images. The distribution of key features varies across query images (for example, portrait retrieval emphasizes texture, while landscape retrieval emphasizes color). The global weight cannot dynamically respond to these differences, resulting in reduced retrieval accuracy.

[0005] Although adaptive weight strategies (such as weight calculation based on feature variance or information entropy) can improve accuracy, they rely on complex matrix operations or iterative optimization, making it difficult to meet real-time retrieval requirements and reducing retrieval efficiency.

[0006] In addition, the single-feature independent weighting strategy ignores the collaborative relationship between multiple features. For example, the joint discriminability of color and spatial features is not effectively utilized, which also reduces the retrieval accuracy. Summary of the Invention

[0007] In order to solve the technical problems existing in the above-mentioned background technology, the present invention provides a method and system for determining feature weights for feature fusion image retrieval. The present invention improves the accuracy of feature fusion image retrieval by selecting the optimal weights of fusion features.

[0008] In order to achieve the above object, the present invention adopts the following technical solutions:

[0009] A first aspect of the present invention provides a method for determining feature weights for feature fusion image retrieval.

[0010] A method for determining feature weights for feature fusion image retrieval, comprising:

[0011] Obtain the single feature metric of each image, construct a single feature metric set, and perform normalization to obtain the normalized single feature metric and single feature metric set;

[0012] Based on the normalized single feature metric, the degree of dispersion and uncertainty are calculated; based on the degree of dispersion and uncertainty, the single feature metric evaluation value is calculated to construct an evaluation set of all single feature metric evaluation values;

[0013] Calculate the correlation matrix based on the correlation between elements in the normalized single-feature metric set;

[0014] Select the optimal index corresponding to the optimal evaluation from the evaluation set, find the optimal metric in the single-feature metric set based on the optimal index, and classify the optimal metric into the applicable metric set, and the optimal index into the applicable index set; define the benchmark index, find the column and row corresponding to the minimum correlation in the correlation matrix, and record it as the matching index; find the matching metric in the single-feature metric set and the matching evaluation in the evaluation set based on the matching index; construct the applicable metric set based on the matching evaluation and matching metric, and determine the weight of the applicable metric and the weight of the redundant metric.

[0015] Furthermore, the degree of dispersion is calculated using the following formula:

[0016]

[0017] in, Indicates the degree of discreteness, Representation of metrics The corresponding normalized single-feature metric The value of the j-th dimension of , N represents the total number of images.

[0018] Furthermore, the uncertainty is calculated using the following formula:

[0019]

[0020] in, Indicates uncertainty, Representation of metrics The corresponding normalized single-feature metric The value of the j-th dimension of .

[0021] Furthermore, the single feature metric evaluation value is calculated using the following formula:

[0022]

[0023] in, represents the single feature metric evaluation value, Indicates the degree of discreteness, Indicates uncertainty.

[0024] Furthermore, the benchmark index is defined, and the column and row corresponding to the minimum correlation are found in the correlation matrix and recorded as the matching index; the method includes:

[0025] The optimal index added is the benchmark index, the corresponding measurement is the benchmark measurement, and the corresponding evaluation is the benchmark evaluation;

[0026] In the benchmark index row of the correlation matrix, after excluding the column corresponding to the index in the applicable index set, find the minimum correlation and obtain the column number corresponding to the minimum correlation, which is recorded as the matching index.

[0027] Furthermore, the method of constructing an applicable metric set based on the matching evaluation and the matching metric, and determining the weights of the applicable metric and the weights of the redundant metric, includes:

[0028] Calculate the absolute value of the difference between the benchmark evaluation and the matching evaluation. If the absolute value is greater than the set threshold, the matching metric is classified into the applicable metric set and the matching index is classified into the applicable index set.

[0029] The weight of the relevant metric in the redundant metric set is set to 0, which is the weight of the redundant metric;

[0030] The corresponding evaluations extracted from the evaluation set by the applicable index set are normalized to obtain the weights of the applicable metrics.

[0031] A second aspect of the present invention provides a feature weight determination system for feature fusion image retrieval.

[0032] A feature weight determination system for feature fusion image retrieval, comprising:

[0033] A normalization processing module is configured to: obtain a single feature metric of each image, construct a single feature metric set, and perform normalization processing to obtain a normalized single feature metric and a single feature metric set;

[0034] An evaluation set construction module is configured to: calculate the degree of dispersion and uncertainty based on the normalized single feature metric; calculate the single feature metric evaluation value based on the degree of dispersion and uncertainty to construct an evaluation set of all single feature metric evaluation values;

[0035] A correlation matrix module is configured to: calculate a correlation matrix based on the correlation between elements in the normalized single feature metric set;

[0036] The weight calculation and retrieval module is configured to: select the optimal index corresponding to the optimal evaluation from the evaluation set, search for the optimal metric in the single-feature metric set based on the optimal index, classify the optimal metric into the applicable metric set, and classify the optimal index into the applicable index set; define the benchmark index, search for the column and row corresponding to the minimum correlation in the correlation matrix, and record it as the matching index; search for the matching metric in the single-feature metric set and the matching evaluation in the evaluation set based on the matching index; construct the applicable metric set based on the matching evaluation and the matching metric, and determine the weight of the applicable metric and the weight of the redundant metric.

[0037] A third aspect of the present invention provides a computer device, comprising:

[0038] a processor adapted to execute a computer program;

[0039] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps in the feature weight determination method for feature fusion image retrieval as described in the first aspect above are implemented.

[0040] The fourth aspect of the present invention provides a computer-readable storage medium storing a computer program, which is suitable for being loaded by a processor and executing the steps in the feature weight determination method for feature fusion image retrieval as described in the first aspect above.

[0041] A fifth aspect of the present invention provides a computer program product or computer program.

[0042] The present invention provides a computer program product or computer program, comprising computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the steps of the method for determining feature weights for feature-fused image retrieval as described in the first aspect above.

[0043] Compared with the prior art, the present invention has the following beneficial effects:

[0044] The existing global fixed weight strategy cannot adapt to the different characteristics of different query images, and the adaptive weights are often compared, which often have lower accuracy. The existing adaptive methods often obtain single feature weights through complex calculations, which significantly reduces the retrieval efficiency and other technical problems. The present invention provides a feature weight determination method and system for feature fusion image retrieval. After obtaining the measurement of a single feature, the evaluation value of the single feature for the data description is first calculated, and then the correlation between different feature measurements is calculated based on the correlation entropy. Finally, based on the correlation and the evaluation value, redundant features and applicable features are obtained, and the weights of the redundant features and the weights of the applicable features are calculated. The comprehensive similarity measurement is calculated, and the comprehensive similarity measurement is sorted. The image retrieval results are returned according to the similarity from large to small. The algorithm of the present invention adaptively determines the weights, adapts to different query image characteristics, is simple to calculate, has better accuracy than the average weight, and is more efficient than the existing adaptive weight determination technology, which can effectively improve the retrieval efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] The accompanying drawings, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0046] Figure 1 is a flow chart of a method for determining feature weights for feature fusion image retrieval according to an embodiment of the present invention;

[0047] Figure 2 is a flow chart of another embodiment of a method for determining feature weights for feature fusion image retrieval according to an embodiment of the present invention;

[0048] Figure 3 is a flow chart showing the calculation of evaluation values ​​according to an embodiment of the present invention;

[0049] Figure 4 is a flow chart of calculating a correlation matrix according to an embodiment of the present invention;

[0050] Figure 5 is a structural diagram of a feature weight determination system for feature fusion image retrieval according to an embodiment of the present invention;

[0051] Figure 6 It is a structural diagram of a computer device shown in an embodiment of the present invention. DETAILED DESCRIPTION

[0052] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0053] It should be noted that the following detailed descriptions are illustrative and intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which the present invention belongs.

[0054] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, they indicate the presence of features, steps, operations, devices, components and / or combinations thereof.

[0055] Figure 1 is a flow chart of a method for determining feature weights for feature fusion image retrieval according to an embodiment of the present invention; Figure 1 ,This method can be applied to medical image retrieval, remote sensing image retrieval, mural image retrieval, etc., including:

[0056] Obtain the single feature metric of each image, construct a single feature metric set, and perform normalization to obtain the normalized single feature metric and single feature metric set;

[0057] Based on the normalized single feature metric, the degree of dispersion and uncertainty are calculated; based on the degree of dispersion and uncertainty, the single feature metric evaluation value is calculated to construct an evaluation set of all single feature metric evaluation values;

[0058] Calculate the correlation matrix based on the correlation between elements in the normalized single-feature metric set;

[0059] Select the optimal index corresponding to the optimal evaluation from the evaluation set, find the optimal metric in the single-feature metric set based on the optimal index, and classify the optimal metric into the applicable metric set, and the optimal index into the applicable index set; define the benchmark index, find the column and row corresponding to the minimum correlation in the correlation matrix, and record it as the matching index; find the matching metric in the single-feature metric set and the matching evaluation in the evaluation set based on the matching index; construct the applicable metric set based on the matching evaluation and matching metric, and determine the weight of the applicable metric and the weight of the redundant metric.

[0060] During the database establishment phase, the present invention collects image data to form an image database and extracts several single features (such as image color features and image texture features) from each image to form an image feature library. During the retrieval phase, after the user enters the image to be retrieved, several single features of the image to be retrieved are extracted. Based on each single feature, the similarity between the image to be retrieved and the images in the database is obtained, forming a single feature metric. All single feature metrics constitute a single feature metric set. The method of the present invention is then invoked to obtain single feature weights based on the single feature metrics, thereby calculating a comprehensive similarity metric: the metric corresponding to the redundant feature × the weight of the redundant feature + the metric corresponding to the applicable feature × the weight of the applicable feature. Finally, the comprehensive similarity metrics are sorted and the image retrieval results are returned in descending order of similarity.

[0061] Figure 2 is a flow chart of another embodiment of a method for determining feature weights for feature fusion image retrieval according to an embodiment of the present invention; Figure 2 , the specific implementation process of this method can be:

[0062] Assuming that there are K features involved in fusion, for a single retrieval, Represents the metric obtained based on the i-th feature, and the single feature metric set is recorded as , Indicates the weight based on the i-th feature, first calculate the single feature metric The assessed value and single feature metric sets The correlation matrix of all metrics in , then based on the evaluation value and correlation matrix Redundant metrics and applicable metrics are distinguished, and finally the weights of redundant metrics and applicable metrics are calculated respectively to obtain the final weights.

[0063] Figure 3 is a flowchart of calculating the evaluation value shown in an embodiment of the present invention; Figure 3 , assuming there are N images in the database, the single feature metric is expressed as In this embodiment, the single-feature metric refers to the vector formed by calculating the Euclidean distance between the image to be retrieved and each image in the database based on a single feature; first, the metric is normalized to obtain the normalized metric As shown in formula (1):

[0064] (1)

[0065] Then the degree of dispersion and uncertainty are calculated as shown in formula (2) and formula (3) respectively.

[0066] (2)

[0067] (3)

[0068] Finally, the single feature metric evaluation value is calculated as shown in formula (3).

[0069] (4)

[0070] The higher the single feature metric evaluation value, the more suitable the feature is for retrieval tasks. Calculate the evaluation values ​​of all single feature metrics and obtain the evaluation set as .

[0071] Figure 4 is a flow chart showing the calculation of the correlation matrix according to an embodiment of the present invention; Figure 4 , traverse each metric in the single feature metric set , normalize it using formula (1) to obtain the normalized metric , construct a normalized single feature metric set . Correlation Matrix The calculation of the element in the i-th row and s-th column in is shown in formula (5).

[0072] (5)

[0073] here, Indicates the number of index features, Indicates the number of index features, represents the index feature dimension, Representation of metrics The corresponding normalized metric The value of the j-th dimension of Representation of metrics The corresponding normalized metric The value of the j-th dimension of Representation of metrics and The greater the correlation, the greater the value.

[0074] In this embodiment, based on the evaluation value and the correlation matrix, the applicable metric set SU and the redundant metric set RE are divided. The algorithm is as follows:

[0075] Input: Single feature metric set , evaluation set , the correlation matrix

[0076] Output: applicable metric set SU, redundant metric set

[0077] The specific process is as follows:

[0078] Step 0: Initialize the threshold , , , applicable index set

[0079] Step 1: First, select the optimal evaluation value from the evaluation set to obtain the corresponding optimal index (the optimal index refers to the position of the optimal evaluation value in the evaluation set PG), find the optimal metric in the single-feature metric set based on the optimal index, and classify it into the applicable metric set. , the optimal index is included in the applicable index set For example, in PG, P5 is the largest, so the optimal index is 5.

[0080] Step 2: The labels of the metrics in the applicable metric set are included in the previous step as benchmark indices. The corresponding metrics are called benchmark metrics, and the corresponding evaluations are called benchmark evaluations. In the row of the correlation matrix containing the benchmark index, after excluding the columns corresponding to the indices in the applicable index set, find the minimum correlation. The column number corresponding to the minimum correlation is obtained and recorded as the matching index. For example, if the optimal index found in the previous step is 5, the row containing the benchmark index here refers to row 5 of the correlation matrix.

[0081] Step 3: According to the matching index, find the corresponding metric in the single-feature metric set and record it as the matching metric, and find the corresponding evaluation in the evaluation set and record it as the matching evaluation.

[0082] Step 4: Calculate the absolute value of the difference between the benchmark evaluation and the matching evaluation. If the absolute value is greater than the threshold, the matching metric is classified into the applicable metric set. , matching indexes are included in the applicable index set , go to Step 2. Otherwise, the algorithm terminates.

[0083] The weight of the relevant metrics in the redundant metric set RE is 0. The weight of the relevant metrics in the applicable metric set SU is obtained according to its corresponding evaluation, that is, based on the applicable index set In the evaluation set After extracting the corresponding evaluation, the evaluation is normalized to obtain the weight. In practical applications, the weights of the relevant metrics in the redundant metric set and the weights of the relevant metrics in the applicable metric set are different when different images to be retrieved are input.

[0084] In this embodiment, based on the applicable index set In the evaluation set Extracting the corresponding evaluations can be understood as: the evaluation set has k elements. The applicable index set SI contains the position of the metrics in the applicable metric set SU in the metric set DD, which is also the position in the evaluation set PG. For example, if SI = {1, 5, 7}, "extracting the corresponding evaluations" means extracting {P1, P5, P7} from PG.

[0085] Aiming at the task of feature fusion image retrieval, this paper proposes a new weight calculation method, which can adaptively search for images, is simple to calculate, has better accuracy than average weight, and is more efficient than existing adaptive weight determination technologies.

[0086] The present invention takes remote sensing image retrieval as an example, and the method according to the present invention includes:

[0087] (1) In the feature library construction phase, 12-band high-resolution satellite images were collected and five types of features were extracted: spectral features (HSV color histogram), texture features (GLCM energy value), spatial features (SIFT key points), vegetation index (NDVI value) and geometric features (shape compactness);

[0088] (2) In the retrieval stage, the user inputs the remote sensing image to be retrieved and extracts the single features of the remote sensing image to be retrieved: spectral features, texture features, spatial features, vegetation index and geometric features; based on each single feature, the similarity between the remote sensing image to be retrieved and the remote sensing images in the database can be obtained to form a single feature metric to construct a single feature metric set, and normalization is performed to obtain the normalized single feature metric and single feature metric set; based on the normalized single feature metric, the degree of dispersion and uncertainty are calculated; based on the degree of dispersion and uncertainty, the single feature metric evaluation value is calculated to construct an evaluation set of all single feature metric evaluation values; based on the correlation between the elements in the normalized single feature metric set, the correlation matrix is ​​calculated; the best single feature metric is selected from the evaluation set. The optimal index corresponding to the optimal evaluation is found in the single-feature metric set according to the optimal index, and the optimal metric is included in the applicable metric set, and the optimal index is included in the applicable index set; the benchmark index is defined, and the column and row corresponding to the minimum correlation in the correlation matrix are found, which is recorded as the matching index; according to the matching index, the matching metric is found in the single-feature metric set, and the matching evaluation is found in the evaluation set; according to the matching evaluation and matching metric, the applicable metric set is constructed, and the weight of the applicable metric and the weight of the redundant metric are determined; the comprehensive similarity metric is calculated = the metric corresponding to the redundant feature × the weight of the redundant feature + the metric corresponding to the applicable feature × the weight of the applicable feature, and finally the comprehensive similarity metric is sorted, and the remote sensing image retrieval results are returned according to the similarity from large to small.

[0089] The present invention can be applied to various image retrieval scenarios, and the above-mentioned remote sensing image retrieval application examples should not be understood as limiting the present invention.

[0090] Combination of the above Figure 1 The feature weight determination method for feature fusion image retrieval provided by an embodiment of the present invention is introduced in detail. Next, the feature weight determination system for feature fusion image retrieval provided by an embodiment of the present invention will be introduced with reference to the accompanying drawings.

[0091] Figure 5 This is a structural diagram of a feature weight determination system for feature fusion image retrieval according to an embodiment of the present invention, referring to Figure 5 , the system of the present invention comprises:

[0092] A normalization processing module is configured to: obtain a single feature metric of each image, construct a single feature metric set, and perform normalization processing to obtain a normalized single feature metric and a single feature metric set;

[0093] An evaluation set construction module is configured to: calculate the degree of dispersion and uncertainty based on the normalized single feature metric; calculate the single feature metric evaluation value based on the degree of dispersion and uncertainty to construct an evaluation set of all single feature metric evaluation values;

[0094] A correlation matrix module is configured to: calculate a correlation matrix based on the correlation between elements in the normalized single feature metric set;

[0095] The weight calculation and retrieval module is configured to: select the optimal index corresponding to the optimal evaluation from the evaluation set, search for the optimal metric in the single-feature metric set based on the optimal index, classify the optimal metric into the applicable metric set, and classify the optimal index into the applicable index set; define the benchmark index, search for the column and row corresponding to the minimum correlation in the correlation matrix, and record it as the matching index; search for the matching metric in the single-feature metric set and the matching evaluation in the evaluation set based on the matching index; construct the applicable metric set based on the matching evaluation and the matching metric, and determine the weight of the applicable metric and the weight of the redundant metric.

[0096] In some embodiments, the discrete degree is calculated using the following formula:

[0097]

[0098] in, Indicates the degree of discreteness, Representation of metrics The corresponding normalized single-feature metric The value of the j-th dimension of , N represents the total number of images.

[0099] In some embodiments, the uncertainty is calculated using the following formula:

[0100]

[0101] in, Indicates uncertainty, Representation of metrics The corresponding normalized single-feature metric The value of the j-th dimension of .

[0102] In some embodiments, the single feature metric evaluation value is calculated using the following formula:

[0103]

[0104] in, represents the single feature metric evaluation value, Indicates the degree of discreteness, Indicates uncertainty.

[0105] In some embodiments, defining a benchmark index and searching for a column and row corresponding to a minimum correlation in a correlation matrix, which is recorded as a matching index, includes:

[0106] The optimal index added is the benchmark index, the corresponding measurement is the benchmark measurement, and the corresponding evaluation is the benchmark evaluation;

[0107] In the benchmark index row of the correlation matrix, after excluding the column corresponding to the index in the applicable index set, find the minimum correlation and obtain the column number corresponding to the minimum correlation, which is recorded as the matching index.

[0108] In some embodiments, constructing an applicable metric set based on the matching evaluation and the matching metric, and determining weights of the applicable metrics and weights of the redundant metrics, comprises:

[0109] Calculate the absolute value of the difference between the benchmark evaluation and the matching evaluation. If the absolute value is greater than the set threshold, the matching metric is classified into the applicable metric set and the matching index is classified into the applicable index set.

[0110] The weight of the relevant metric in the redundant metric set is set to 0, which is the weight of the redundant metric;

[0111] The corresponding evaluations extracted from the evaluation set by the applicable index set are normalized to obtain the weights of the applicable metrics.

[0112] The feature weight determination system for feature fusion image retrieval according to an embodiment of the present invention may correspond to the method described in the embodiment of the present invention, and the above and other operations and / or functions of each module of the feature weight determination system for feature fusion image retrieval are respectively to achieve Figure 1 For the sake of brevity, the corresponding processes of each method in are not repeated here.

[0113] See also Figure 6 The structural diagram of the computer device shown in the figure includes a processor, a communication interface and a computer-readable storage medium. The processor, the communication interface and the computer-readable storage medium can be connected via a bus or other means. The communication interface is used to receive and send data. The computer-readable storage medium can be stored in the memory of the computer device, and the computer-readable storage medium is used to store a computer program. The computer program includes program instructions, and the processor is used to execute the program instructions stored in the computer-readable storage medium. The processor (or CPU (Central Processing Unit)) is the computing core and control core of the computer device, which is suitable for implementing one or more instructions, and is specifically suitable for loading and executing one or more instructions to implement the corresponding steps in the embodiment of the feature weight determination method for feature fusion image retrieval.

[0114] This embodiment provides a computer-readable storage medium (Memory). This computer-readable storage medium is a memory device within a computer device that is used to store programs and data. It should be understood that the computer-readable storage medium herein may include both built-in storage media within the computer device and, of course, extended storage media supported by the computer device. The computer-readable storage medium provides storage space that stores the processing system of the computer device.

[0115] Furthermore, the storage space also stores one or more instructions suitable for being loaded and executed by the processor. These instructions may be one or more computer programs (including program code). It should be noted that the computer-readable storage medium herein may be a high-speed RAM memory or a non-volatile memory, such as at least one disk storage device; alternatively, it may be at least one computer-readable storage medium located remotely from the processor.

[0116] In one embodiment, one or more instructions are stored in the computer-readable storage medium; the processor loads and executes the one or more instructions stored in the computer-readable storage medium to implement the corresponding steps in the above-mentioned embodiment of the feature weight determination method for feature fusion image retrieval.

[0117] This embodiment provides a computer program product or computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the corresponding steps of the above-mentioned embodiment of the method for determining feature weights for feature fusion image retrieval.

[0118] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of hardware embodiments, software embodiments, or embodiments combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage and optical storage, etc.) containing computer-usable program code.

[0119] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0120] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0121] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0122] Those skilled in the art will appreciate that all or part of the processes in the above-described method embodiments can be implemented by instructing the relevant hardware through a computer program. The program can be stored in a computer-readable storage medium. When executed, the program can include the processes in the above-described method embodiments. The storage medium can be a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0123] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention shall be included within the scope of protection of the present invention.

Claims

1. A method for determining feature weights for feature fusion image retrieval, characterized in that: include: Obtain the single feature metric of each image, construct a single feature metric set, and perform normalization to obtain the normalized single feature metric and single feature metric set; Calculate the degree of dispersion and uncertainty based on the normalized single feature metric; Based on the degree of dispersion and uncertainty, the single feature metric evaluation value is calculated to construct an evaluation set of all single feature metric evaluation values; Calculate the correlation matrix based on the correlation between elements in the normalized single-feature metric set; Select the optimal index corresponding to the optimal evaluation from the evaluation set, find the optimal metric in the single-feature metric set based on the optimal index, and put the optimal metric into the applicable metric set and the optimal index into the applicable index set; Define the benchmark index and find the column and row corresponding to the minimum correlation in the correlation matrix, which is recorded as the matching index; According to the matching index, find the matching metric in the single feature metric set and find the matching evaluation in the evaluation set; According to the matching evaluation and matching measurement, an applicable measurement set is constructed, and the weights of the applicable measurements and the weights of the redundant measurements are determined; The optimal index added is the benchmark index, the corresponding metric is the benchmark metric, and the corresponding evaluation is the benchmark evaluation. In the benchmark index row of the correlation matrix, after excluding the column corresponding to the index in the applicable index set, find the minimum correlation and obtain the column number corresponding to the minimum correlation, which is recorded as the matching index. Calculate the absolute value of the difference between the benchmark evaluation and the matching evaluation. If the absolute value is greater than the set threshold, the matching metric is classified into the applicable metric set and the matching index is classified into the applicable index set. The weight of the relevant metric in the redundant metric set is set to 0, which is the weight of the redundant metric. The corresponding evaluations extracted from the evaluation set by the applicable index set are normalized to obtain the weights of the applicable metrics.

2. The method for determining feature weights for feature fusion image retrieval according to claim 1, wherein: The degree of dispersion is calculated using the following formula: in, Indicates the degree of discreteness, Representation of metrics The corresponding normalized single-feature metric The value of the j-th dimension of , N represents the total number of images.

3. The method for determining feature weights for feature fusion image retrieval according to claim 1, wherein: The uncertainty is calculated using the following formula: in, Indicates uncertainty, Representation of metrics The corresponding normalized single-feature metric The value of the j-th dimension of .

4. The method for determining feature weights for feature fusion image retrieval according to claim 1, wherein: The single feature metric evaluation value is calculated using the following formula: in, represents the single feature metric evaluation value, Indicates the degree of discreteness, Indicates uncertainty.

5. A feature weight determination system for feature fusion image retrieval, characterized in that: include: A normalization processing module is configured to: obtain a single feature metric of each image, construct a single feature metric set, and perform normalization processing to obtain a normalized single feature metric and a single feature metric set; The evaluation set construction module is configured to: calculate the degree of dispersion and uncertainty based on the normalized single-feature metric; Based on the degree of dispersion and uncertainty, the single feature metric evaluation value is calculated to construct an evaluation set of all single feature metric evaluation values; A correlation matrix module is configured to: calculate a correlation matrix based on the correlation between elements in the normalized single feature metric set; a weight calculation and retrieval module configured to: select an optimal index corresponding to the optimal evaluation from the evaluation set, search for an optimal metric in the single-feature metric set based on the optimal index, assign the optimal metric to the applicable metric set, and assign the optimal index to the applicable index set; Define the benchmark index and find the column and row corresponding to the minimum correlation in the correlation matrix, which is recorded as the matching index; According to the matching index, find the matching metric in the single feature metric set and find the matching evaluation in the evaluation set; According to the matching evaluation and matching measurement, an applicable measurement set is constructed, and the weights of the applicable measurements and the weights of the redundant measurements are determined; The optimal index added is the benchmark index, the corresponding metric is the benchmark metric, and the corresponding evaluation is the benchmark evaluation. In the benchmark index row of the correlation matrix, after excluding the column corresponding to the index in the applicable index set, find the minimum correlation and obtain the column number corresponding to the minimum correlation, which is recorded as the matching index. Calculate the absolute value of the difference between the benchmark evaluation and the matching evaluation. If the absolute value is greater than the set threshold, the matching metric is classified into the applicable metric set and the matching index is classified into the applicable index set. The weight of the relevant metric in the redundant metric set is set to 0, which is the weight of the redundant metric. The corresponding evaluations extracted from the evaluation set by the applicable index set are normalized to obtain the weights of the applicable metrics.

6. A computer device, characterized in that: a processor adapted to execute a computer program; A computer-readable storage medium having a computer program stored therein, wherein the computer program, when executed by the processor, implements the steps in the method for determining feature weights for feature fusion image retrieval according to any one of claims 1 to 4.

7. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, which is suitable for being loaded by a processor and executing the steps in the feature weight determination method for feature fusion image retrieval according to any one of claims 1 to 4.

8. A computer program product, characterized in that The computer program product includes a computer program, and when the computer program is executed by a processor, the computer program implements the steps in the method for determining feature weights for feature fusion image retrieval according to any one of claims 1 to 4.

Images