CN104850861B - Based on the fungal keratitis image-recognizing method of RX abnormality detection and texture analysis - Google Patents
Based on the fungal keratitis image-recognizing method of RX abnormality detection and texture analysis Download PDFInfo
- Publication number
- CN104850861B CN104850861B CN201510278952.4A CN201510278952A CN104850861B CN 104850861 B CN104850861 B CN 104850861B CN 201510278952 A CN201510278952 A CN 201510278952A CN 104850861 B CN104850861 B CN 104850861B
- Authority
- CN
- China
- Prior art keywords
- training sample
- mycelia
- image
- neural
- test sample
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
- G06F18/24133—Distances to prototypes
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Eye Examination Apparatus (AREA)
Abstract
The invention discloses the fungal keratitis image-recognizing method based on RX abnormality detection and texture analysis, comprising: obtain normal cornea neuro images and only comprise mycelia mycelia image as training sample; Obtain the RETINAL IMAGES of fungal keratitis patient as test sample book; Pre-service, feature extraction and Fusion Features are carried out to the normal cornea neuro images in training sample, obtains the Neuronal Characteristics after training sample fusion; Pre-service, feature extraction and Fusion Features are carried out to the mycelia image only comprising mycelia in training sample, obtains the mycelia feature after training sample fusion; Pre-service, feature extraction and Fusion Features are carried out to the image in test sample book, obtains the mycelia feature after the Neuronal Characteristics after test sample book fusion and test sample book fusion; Identify the nerve in test sample book and mycelia.
Description
Technical field
The invention belongs to technical field of image processing, particularly relate to a kind of fungal keratitis image-recognizing method based on RX abnormality detection and texture analysis.
Background technology
Fungal keratitis is the high infectious disease of cornea of a kind of blind rate caused by pathomycete, owing to developing rapidly, very easily cause the serious consequences such as perforation of cornea, hypopyon, entophthamia, thus endanger extremely heavy, the early diagnosis and therapy of this disease is seemed particularly important.Confocal microscope is a kind of novel, non-invasive cornea imaging examination instrument, it corneal can carry out scanning imagery from four-dimensional (three dimensions and time) level on live body, and provide the cornea of high definition and enlargement ratio each layer images, make people from cellular level, direct observational study can be carried out to the pathologic, physiologic of live cornea.
Confocal microscope obtains image method: suffer from eye 0.4% oxybuprocaine drop eye 2 times, eyelid left by eye speculum, and the lower jaw of patient and forehead are fixed on the inspection bracket on microscope.The artificial tears (Vidisic-Gel) waiting and open is coated with on the 40 times of coniform object lens of immersion type (achroplan400.75W) surfaces, by light and slow for camera lens reach, the artificial tears on surface is contacted with keratopathy place, and the distance between camera lens and cornea is 2mm.Slowly moved forward by camera lens, namely the scan image of each layer of cornea shows fast by computer monitor, and image is simultaneously by S-VHS video recorder record, and check and terminate to put video recording slowly, careful story board rechecking afterwards, selected image comparatively is clearly stored in computing machine.
In the eye fundus image of normal cornea, only has neural existence, so we diagnose fungal keratitis mainly to observe mycelia for foundation with confocal microscope.In the eye fundus image of fungal keratitis patient, mycelia and corneal stroma nerve are common existence, and straight long wire mycelia is easy to obscure mutually with corneal stroma nerve.Mycelia is in brighter thread-like morphology under relatively dark background, and the separation of most visible mycelia, fills the air distribution, be interweaved, the Substance P under corneal epithelium has certain rule, many in trident or Y shape branch, its direction out of shape is consistent, orderly, is not formed interlaced between nerve.
The image of current collection comprises simple mycelia image, normal cornea neuro images, and the eye fundus image of fungal keratitis patient.A large amount of medical images causes two problems: on the one hand, and the inspection image reading patient becomes a very hard work of doctor, due to long-time interpretation image, makes the easy fatigue of doctor and diverts one's attention, thus causing the decline of rate of correct diagnosis; On the other hand, only rely on the experience of doctor self to be difficult to ensure there will not be the situation of failing to pinpoint a disease in diagnosis with mistaken diagnosis, and be difficult to carry out consistent quantitative test to image data, and be the inevitable requirement of Medical Imaging development to the quantitative test of medical image.In this context, Medical Image Processing and the analytical technology status in Medical Imaging just seems more and more important.So-called Medical Image Processing and analysis are exactly this instrument of computer, utilize the method for mathematics to carry out various processing and process to medical image, to provide more diagnostic message or data for clinical according to clinical specifically needs.Such as, for the image that contrast is undesirable and signal to noise ratio (S/N ratio) is not high, utilize the method for image enhaucament and filtering to change the contrast of image, improve the signal to noise ratio (S/N ratio) of image, thus be supplied to the image of doctor's better quality, so that doctor is to the interpretation of image; For a large amount of medical images that the medical imaging device by advanced person produces, after first can carrying out image procossing by computing machine, suspicious focus is all marked, and then by doctor, interpretation is carried out to the suspicious lesions be marked.The read tablet time that radiologist is a large amount of can be saved like this, them are made to be able to a visual cognitive ability on suspicious lesions, thus lay the foundation for correctly diagnosing, Here it is obtains extensive concern in Medical Imaging field at present and develops the technology of medical image computer-aided diagnosis rapidly.
In the example of fungal keratitis, how to utilize simple mycelia image and neuro images, extract mycelia respectively and neural feature carrys out training classifier, realizing mycelia in fungal keratitis patient eye fundus image and neural classification is a main task.In the image obtained, because the interference of background is very serious, directly can not carries out effective feature extraction, therefore need to carry out pre-service to image.Traditional image pre-processing method such as luminance transformation, spatial filtering, the performance of the methods such as histogram treatment is not very well, can not the interference of highly effective removal background.How effectively to remove background interference and to utilize the difference on mycelia and neuronal nitric-oxide synthase to be good problems to study to distinguish them.
Summary of the invention
Object of the present invention is exactly to solve the problem, and provide a kind of fungal keratitis image-recognizing method based on RX abnormality detection and texture analysis, it has the high advantage of discrimination.
To achieve these goals, the present invention adopts following technical scheme:
Based on the fungal keratitis image-recognizing method of RX abnormality detection and texture analysis, comprising:
Step (1): obtain normal cornea neuro images and only comprise mycelia mycelia image as training sample; Obtain the RETINAL IMAGES of fungal keratitis patient as test sample book;
Step (2): described step (2) comprises concurrent step (2-1), step (2-2) and step (2-3) of carrying out;
Step (2-1): carry out pre-service, feature extraction and Fusion Features to the normal cornea neuro images in training sample, obtains the Neuronal Characteristics after training sample fusion;
Step (2-2): carry out pre-service, feature extraction and Fusion Features to the mycelia image only comprising mycelia in training sample, obtains the mycelia feature after training sample fusion;
Step (2-3): carry out pre-service, feature extraction and Fusion Features to test sample book, obtains the mycelia feature after the Neuronal Characteristics after test sample book fusion and test sample book fusion;
Step (3):
Euclidean distance between Neuronal Characteristics after Neuronal Characteristics after being merged by calculating test sample book and training sample are merged, identifies the nerve in test sample book;
Euclidean distance between mycelia feature after mycelia feature after being merged by calculating test sample book and training sample are merged, identifies the mycelia in test sample book.
The step of described step (2-1) is as follows:
RX abnormality detection algorithm is adopted to carry out pre-service to the normal cornea neuro images in training sample, the neural textural characteristics of neural shapes characteristic sum training sample of the neural length characteristic of training sample, training sample is extracted respectively from pretreated training sample, and the neural textural characteristics of the neural shapes characteristic sum training sample of the neural length characteristic of training sample, training sample is carried out Fusion Features, obtain the Neuronal Characteristics after training sample fusion.
The step of described step (2-2) is as follows:
RX abnormality detection algorithm is adopted to carry out pre-service to the mycelia image only comprising mycelia in training sample, the mycelia textural characteristics of the Hyphal length feature of training sample, the mycelia shape facility of training sample and training sample is extracted respectively from pretreated training sample, and the mycelia textural characteristics of the Hyphal length feature of training sample, the mycelia shape facility of training sample and training sample is carried out Fusion Features, obtain the mycelia feature after training sample fusion.
The step of described step (2-3) is as follows:
Adopt the test sample book of RX abnormality detection algorithm to step (1) to carry out pre-service, from pretreated test sample book, extract the mycelia textural characteristics of the Hyphal length feature of test sample book, the neural length characteristic of test sample book, the neural shapes feature of test sample book, the mycelia shape facility of test sample book, the neural textural characteristics of test sample book and test sample book respectively;
Fusion Features is carried out to the neural textural characteristics of the neural length characteristic of test sample book extracted, the neural shapes characteristic sum test sample book of test sample book, obtains the Neuronal Characteristics after test sample book fusion;
Fusion Features is carried out to the mycelia textural characteristics of the Hyphal length feature of test sample book extracted, the mycelia shape facility of test sample book and test sample book, obtains the mycelia feature after test sample book fusion.
The detailed process of described step (2-1) is:
Step (2-1-1): adopt RX abnormality detection algorithm to carry out pre-service to the normal cornea neuro images in training sample, remove the background interference information in training sample;
Step (2-1-2): carry out binary conversion treatment to pretreated training sample, carries out the process of dilation erosion to the image after binary conversion treatment, to the neural length characteristic of the image zooming-out after dilation erosion process;
Step (2-1-3): the neural shapes feature of the image after the process of extraction step (2-1-2) dilation erosion;
Step (2-1-4): the neural textural characteristics of the image after the process of extraction step (2-1-2) dilation erosion;
Step (2-1-5): Fusion Features is carried out to the neural length characteristic of the training sample extracted, neural shapes feature and neural textural characteristics, obtain training sample merge after Neuronal Characteristics.
The step of described step (2-2) and step (2-3) is the same with step (2-1), repeats no more here.
The step of described step (2-1-2) is as follows:
Step (2-1-2-1): binary conversion treatment is carried out to pretreated training sample, and use the method for connected domain, the threshold value of setting connected domain area, removes the background interference exceeding setting threshold value;
Step (2-1-2-2): dilation erosion process is carried out to the training sample after binary conversion treatment, described dilation erosion process refers to the post-etching that first expands, first expansion post-etching has the effect of minuscule hole and smooth boundary in filler body, after dilation erosion process, obtain the bianry image of better effects if, dilation erosion process can strengthen nerve information;
Step (2-1-2-3): according to the difference on neuromorphic, tagged process is carried out to each connected domain in the training sample of the nerve after step (2-1-2-2) process, and find the barycenter of each connected domain respectively, then find out distance barycenter point farthest in each connected domain respectively, the distance between this point and barycenter is preserved as neural length characteristic.
The detailed process extracting the neural shapes feature of training sample in described step (2-1-3) is:
For the training sample after the process of step (2-1-2) dilation erosion, utilize Fourier transform to extract the shape facility of connected domain, be neural shape facility.
The detailed process extracting the neural textural characteristics of training sample in described step (2-1-4) is:
The gray level image obtained after the process of step (2-1-2) dilation erosion and original image are carried out dot product, after original image dot product, obtains the gray level image of target area; Again by the method for gray level co-occurrence matrixes, neural textural characteristics is extracted to the gray level image of target area.
In described step (2-1-5) to the detailed process that the neural length characteristic extracted, neural shapes feature and neural textural characteristics carry out Fusion Features be:
The neural length characteristic value, neural shapes feature and the neural textural characteristics that extract are arranged in a row vector, as the proper vector of nerve.
In described step (3):
Euclidean distance between Neuronal Characteristics after Neuronal Characteristics after test sample book merges and training sample merge is less, shows that the Neuronal Characteristics of the Neuronal Characteristics of test sample book and training sample is more close, belongs to same class Neuronal Characteristics;
Euclidean distance between mycelia feature after mycelia feature after test sample book merges and training sample merge is less, shows that the mycelia feature of the mycelia feature of test sample book and training sample is more close, belongs to same class Neuronal Characteristics.
Beneficial effect of the present invention:
(1) have employed the eye fundus image of RX abnormality detection algorithm to shooting in EO-1 hyperion in the present invention and carry out pre-service, effectively can remove the interference of background information, be conducive to the carrying out of characteristic extraction step;
(2) present invention employs the algorithm that two kinds of traditional texture are analyzed, extract statistical information and the architectural feature of image respectively, and add length descriptor and Fourier's shape description symbols according to the situation of real image, finally these features are merged, the feature that single algorithm extracts can not be all in representative picture useful information, adopt the method for Fusion Features the useful information in image more can be extracted, thus improve discrimination.
Accompanying drawing explanation
Fig. 1 (a) is original mycelia image;
Fig. 1 (b) is image after image procossing imadjust functional transformation;
Fig. 1 (c) is mycelia image after RX abnormality detection.
Fig. 2 (a) is original neuro images;
Fig. 2 (b) is image after imadjust functional transformation;
Fig. 2 (c) is image after RX abnormality detection algorithm process;
Fig. 3 (a) is to carrying out image after binary conversion treatment through RX abnormality detection mycelia image;
Fig. 3 (b) is setting threshold value, mycelia image after the connected domain that removal area is less;
Fig. 4 is method flow diagram of the present invention.
Embodiment
Below in conjunction with accompanying drawing and embodiment, the invention will be further described.
As shown in Figure 4, based on the fungal keratitis image-recognizing method of RX abnormality detection and texture analysis, comprising:
Step (1): obtain normal cornea neuro images and only comprise mycelia mycelia image as training sample; Obtain the RETINAL IMAGES of fungal keratitis patient as test sample book;
Step (2): described step (2) comprises concurrent step (2-1), step (2-2) and step (2-3) of carrying out;
Step (2-1): pre-service, feature extraction and Fusion Features are carried out to the normal cornea neuro images in training sample and obtains the Neuronal Characteristics after training sample fusion;
Step (2-2): pre-service, feature extraction and Fusion Features are carried out to the mycelia image only comprising mycelia in training sample and obtains the mycelia feature after training sample fusion;
Step (2-3): pre-service, feature extraction and Fusion Features are carried out to test sample book and obtains the mycelia feature after the Neuronal Characteristics after test sample book fusion and test sample book fusion;
Step (3): the Euclidean distance between the Neuronal Characteristics after the Neuronal Characteristics after being merged by calculating test sample book and training sample are merged, identifies the nerve in test sample book; Euclidean distance between mycelia feature after mycelia feature after being merged by calculating test sample book and training sample are merged, identifies the mycelia in test sample book.
The step of described step (2-1) is as follows: adopt RX abnormality detection algorithm to carry out pre-service to the normal cornea neuro images in training sample, the neural textural characteristics of neural shapes characteristic sum training sample of the neural length characteristic of training sample, training sample is extracted respectively from pretreated training sample, and the neural textural characteristics of the neural shapes characteristic sum training sample of the neural length characteristic of training sample, training sample is carried out Fusion Features, obtain the Neuronal Characteristics after training sample fusion.
The step of described step (2-2) is as follows: adopt RX abnormality detection algorithm to carry out pre-service to the mycelia image only comprising mycelia in training sample, the mycelia textural characteristics of the Hyphal length feature of training sample, the mycelia shape facility of training sample and training sample is extracted respectively from pretreated training sample, and the mycelia textural characteristics of the Hyphal length feature of training sample, the mycelia shape facility of training sample and training sample is carried out Fusion Features, obtain the mycelia feature after training sample fusion.
The step of described step (2-3) is as follows: adopt the test sample book of RX abnormality detection algorithm to step (1) to carry out pre-service, extracts the mycelia textural characteristics of the Hyphal length feature of test sample book, the neural length characteristic of test sample book, the neural shapes feature of test sample book, the mycelia shape facility of test sample book, the neural textural characteristics of test sample book and test sample book from pretreated test sample book respectively; Fusion Features is carried out to the neural textural characteristics of the neural length characteristic of test sample book extracted, the neural shapes characteristic sum test sample book of test sample book, obtains the Neuronal Characteristics after test sample book fusion; Fusion Features is carried out to the mycelia textural characteristics of the Hyphal length feature of test sample book extracted, the mycelia shape facility of test sample book and test sample book, obtains the mycelia feature after test sample book fusion.
The detailed process of described step (2-1) is:
Step (2-1-1): adopt RX abnormality detection algorithm to carry out pre-service to the normal cornea neuro images in training sample, remove the background interference information in training sample;
Step (2-1-2): carry out binary conversion treatment to pretreated training sample, carries out the process of dilation erosion to the image after binary conversion treatment, to the neural length characteristic of the image zooming-out after dilation erosion process;
Step (2-1-3): the neural shapes feature of the image after the process of extraction step (2-1-2) dilation erosion;
Step (2-1-4): the neural textural characteristics of the image after the process of extraction step (2-1-2) dilation erosion;
Step (2-1-5): Fusion Features is carried out to the neural length characteristic of the training sample extracted, neural shapes feature and neural textural characteristics, obtain training sample merge after Neuronal Characteristics.
The step of described step (2-1-2) is as follows:
Step (2-1-2-1): binary conversion treatment is carried out to pretreated training sample, and use the method for connected domain, the threshold value of setting connected domain area, removes the background interference exceeding setting threshold value; Nerve is linear, because the sharpness of original image is bad, can there is the background interference of many point-like after binaryzation, and the threshold value of setting area, can remove the background interference of many point-like.
Step (2-1-2-2): dilation erosion process is carried out to the training sample after binary conversion treatment, described dilation erosion process refers to the post-etching that first expands, first expansion post-etching has the effect of minuscule hole and smooth boundary in filler body, after dilation erosion process, obtain the bianry image of better effects if, dilation erosion process can strengthen nerve information;
Step (2-1-2-3): according to the difference on neuromorphic, tagged process is carried out to each connected domain in the training sample of the nerve after step (2-1-2-2) process, and find the barycenter of each connected domain respectively, then find out distance barycenter point farthest in each connected domain respectively, the distance between this point and barycenter is preserved as neural length characteristic.
The detailed process extracting the neural shapes feature of training sample in described step (2-1-3) is: for the training sample after the process of step (2-1-2) dilation erosion, utilize Fourier transform to extract the shape facility of connected domain, be neural shape facility.
The detailed process extracting the neural textural characteristics of training sample in described step (2-1-4) is: the gray level image obtained after the process of step (2-1-2) dilation erosion and original image are carried out dot product, because background pixel value is 0 in bianry image, target area pixel value is 1, after original image dot product, obtain the gray level image of target area; Again by the method for gray level co-occurrence matrixes, neural textural characteristics is extracted to the gray level image of target area.
In described step (2-1-5) to the detailed process that the neural length characteristic extracted, neural shapes feature and neural textural characteristics carry out Fusion Features be: the neural length characteristic value, neural shapes feature and the neural textural characteristics that extract are arranged in a row vector, as the proper vector of nerve.
Comprise much useful information in piece image, and single feature extraction often can not extract useful informations all in image.And adopt the method for Fusion Features, by obtain length, shape and textural characteristics value, being arranged in a row vector carrys out representative image, can extracting mycelia in image and neural information maximization, thus improves discrimination.
The detailed process of described step (2-2) is:
Step (2-2-1): adopt RX abnormality detection algorithm to carry out pre-service to the mycelia image only comprising mycelia in training sample, remove the background interference information in training sample;
Step (2-2-2): carry out binary conversion treatment to pretreated training sample, carries out the process of dilation erosion to the image after binary conversion treatment, to the image zooming-out Hyphal length feature after dilation erosion process;
Step (2-2-3): the mycelia shape facility of the image after the process of extraction step (2-2-2) dilation erosion;
Step (2-2-4): the mycelia textural characteristics of the image after the process of extraction step (2-2-2) dilation erosion;
Step (2-2-5): Fusion Features is carried out to the Hyphal length feature of the training sample extracted, mycelia shape facility and mycelia textural characteristics, obtain training sample merge after mycelia feature.
The step of described step (2-2-2) is as follows:
Step (2-2-2-1): binary conversion treatment is carried out to pretreated training sample, and use the method for connected domain, the threshold value of setting connected domain area, removes the background interference exceeding setting threshold value; As shown in Fig. 3 (b); Mycelia is linear, because the sharpness of original image is bad, can there is the background interference of many point-like after binaryzation, and the threshold value of setting area, can remove the background interference of many point-like.
Step (2-2-2-2): dilation erosion process is carried out to the training sample after binary conversion treatment, described dilation erosion process refers to the post-etching that first expands, first expansion post-etching has the effect of minuscule hole and smooth boundary in filler body, after dilation erosion process, obtain the bianry image of better effects if, dilation erosion process can strengthen mycelia information;
Step (2-2-2-3): according to the difference on hypha form, tagged process is carried out to each connected domain in the training sample of the mycelia after step (2-2-2-2) process, and find the barycenter of each connected domain respectively, then find out distance barycenter point farthest in each connected domain respectively, the distance between this point and barycenter is preserved as Hyphal length feature.
The detailed process extracting the mycelia shape facility of training sample in described step (2-2-3) is:
For the training sample after the process of step (2-2-2) dilation erosion, utilize Fourier transform to extract the shape facility of connected domain, be the shape facility of mycelia.
The detailed process extracting the mycelia textural characteristics of training sample in described step (2-2-4) is: the gray level image obtained after the process of step (2-2-2) dilation erosion and original image are carried out dot product, because background pixel value is 0 in bianry image, target area pixel value is 1, after original image dot product, obtain the gray level image of target area; Again by the method for gray level co-occurrence matrixes, mycelia textural characteristics is extracted to the gray level image of target area.
In described step (2-2-5) to the detailed process that the Hyphal length feature extracted, mycelia shape facility and mycelia textural characteristics carry out Fusion Features be: the Hyphal length eigenwert extracted, mycelia shape facility and mycelia textural characteristics are arranged in a row vector, as the proper vector of mycelia.
The step of described step (2-3) is the same with step (2-1) and step (2-2), repeats no more here.
In described step (3): the value of Euclidean distance is less, show that the Neuronal Characteristics of the Neuronal Characteristics of test sample book and training sample is more close, belong to same class Neuronal Characteristics; The value of Euclidean distance is less, shows that the mycelia feature of the mycelia feature of test sample book and training sample is more close, belongs to same class Neuronal Characteristics.
In the present embodiment, Fig. 1 (a) is depicted as original mycelia image, Fig. 1 (b) is depicted as the image in traditional images process after imadjust functional transformation, and Fig. 1 (c) is depicted as the image using and draw after RX abnormality detection algorithm in EO-1 hyperion.Can find out, RX abnormality detection algorithm achieves better effect, effectively eliminates the interference of more background information, is conducive to the extraction of feature.
Fig. 2 (a) is depicted as original neuro images, the image that Fig. 2 (b) obtains after being depicted as imadjust conversion, and Fig. 2 (c) is depicted as the image using and obtain after RX abnormality detection algorithm in EO-1 hyperion.Can find out, RX abnormality detection algorithm achieves better effect, effectively eliminates the interference of more background information, is conducive to the extraction of feature.
Fig. 3 (a) is depicted as and image after RX abnormality detection is carried out image after binary conversion treatment, the interference having many backgrounds can be seen, setting connected domain area threshold, removes the value connected domain that area is less than setting connected domain area threshold, obtains image shown in Fig. 3 (b).Fig. 3 (b) is found to the barycenter of each connected domain, then distance barycenter point farthest in each connected domain is found out, obtain the length distance between this point and barycenter, and set a threshold value by experiment, connected domain length being less than this threshold value is removed, background interference can be got rid of to greatest extent like this, and qualified length value is preserved as mycelia eigenwert.Identical process is carried out for neuro images, also show that one group of length distance preserves as the eigenwert of nerve.
By reference to the accompanying drawings the specific embodiment of the present invention is described although above-mentioned; but not limiting the scope of the invention; one of ordinary skill in the art should be understood that; on the basis of technical scheme of the present invention, those skilled in the art do not need to pay various amendment or distortion that creative work can make still within protection scope of the present invention.
Claims (9)
1., based on the fungal keratitis image-recognizing method of RX abnormality detection and texture analysis, it is characterized in that, comprising:
Step (1): obtain normal cornea neuro images and only comprise mycelia mycelia image as training sample; Obtain the RETINAL IMAGES of fungal keratitis patient as test sample book;
Step (2): described step (2) comprises concurrent step (2-1), step (2-2) and step (2-3) of carrying out;
Step (2-1): carry out pre-service, feature extraction and Fusion Features to the normal cornea neuro images in training sample, obtains the Neuronal Characteristics after training sample fusion;
The detailed process of described step (2-1) is:
Step (2-1-1): adopt RX abnormality detection algorithm to carry out pre-service to the normal cornea neuro images in training sample, remove the background interference information in training sample;
Step (2-1-2): carry out binary conversion treatment to pretreated training sample, carries out the process of dilation erosion to the image after binary conversion treatment, to the neural length characteristic of the image zooming-out after dilation erosion process;
Step (2-1-3): the neural shapes feature of the image after the process of extraction step (2-1-2) dilation erosion;
Step (2-1-4): the neural textural characteristics of the image after the process of extraction step (2-1-2) dilation erosion;
Step (2-1-5): Fusion Features is carried out to the neural length characteristic of the training sample extracted, neural shapes feature and neural textural characteristics, obtain training sample merge after Neuronal Characteristics;
Step (2-2): carry out pre-service, feature extraction and Fusion Features to the mycelia image only comprising mycelia in training sample, obtains the mycelia feature after training sample fusion;
Step (2-3): to test sample book carry out pre-service, feature extraction and Fusion Features obtain test sample book merge, after Neuronal Characteristics and test sample book merge after mycelia feature;
Step (3):
Euclidean distance between Neuronal Characteristics after Neuronal Characteristics after being merged by calculating test sample book and training sample are merged, identifies the nerve in test sample book;
Euclidean distance between mycelia feature after mycelia feature after being merged by calculating test sample book and training sample are merged, identifies the mycelia in test sample book.
2., as claimed in claim 1 based on the fungal keratitis image-recognizing method of RX abnormality detection and texture analysis, it is characterized in that, the step of described step (2-1) is as follows:
RX abnormality detection algorithm is adopted to carry out pre-service to the normal cornea neuro images in training sample, the neural textural characteristics of neural shapes characteristic sum training sample of the neural length characteristic of training sample, training sample is extracted respectively from pretreated training sample, and the neural textural characteristics of the neural shapes characteristic sum training sample of the neural length characteristic of training sample, training sample is carried out Fusion Features, obtain the Neuronal Characteristics after training sample fusion.
3., as claimed in claim 1 based on the fungal keratitis image-recognizing method of RX abnormality detection and texture analysis, it is characterized in that, the step of described step (2-2) is as follows:
RX abnormality detection algorithm is adopted to carry out pre-service to the mycelia image only comprising mycelia in training sample, the mycelia textural characteristics of the Hyphal length feature of training sample, the mycelia shape facility of training sample and training sample is extracted respectively from pretreated training sample, and the mycelia textural characteristics of the Hyphal length feature of training sample, the mycelia shape facility of training sample and training sample is carried out Fusion Features, obtain the mycelia feature after training sample fusion.
4., as claimed in claim 1 based on the fungal keratitis image-recognizing method of RX abnormality detection and texture analysis, it is characterized in that, the step of described step (2-3) is as follows:
Adopt the test sample book of RX abnormality detection algorithm to step (1) to carry out pre-service, from pretreated test sample book, extract the mycelia textural characteristics of the Hyphal length feature of test sample book, the neural length characteristic of test sample book, the neural shapes feature of test sample book, the mycelia shape facility of test sample book, the neural textural characteristics of test sample book and test sample book respectively;
Fusion Features is carried out to the neural textural characteristics of the neural length characteristic of test sample book extracted, the neural shapes characteristic sum test sample book of test sample book, obtains the Neuronal Characteristics after test sample book fusion;
Fusion Features is carried out to the mycelia textural characteristics of the Hyphal length feature of test sample book extracted, the mycelia shape facility of test sample book and test sample book, obtains the mycelia feature after test sample book fusion.
5., as claimed in claim 1 based on the fungal keratitis image-recognizing method of RX abnormality detection and texture analysis, it is characterized in that, the step of described step (2-1-2) is as follows:
Step (2-1-2-1): binary conversion treatment is carried out to pretreated training sample, and use the method for connected domain, the threshold value of setting connected domain area, removes the background interference exceeding setting threshold value;
Step (2-1-2-2): dilation erosion process is carried out to the training sample after binary conversion treatment, described dilation erosion process refers to the post-etching that first expands, first expansion post-etching has the effect of minuscule hole and smooth boundary in filler body, after dilation erosion process, obtain the bianry image of better effects if, dilation erosion process can strengthen nerve information;
Step (2-1-2-3): according to the difference on neuromorphic, tagged process is carried out to each connected domain in the training sample of the nerve after step (2-1-2-2) process, and find the barycenter of each connected domain respectively, then find out distance barycenter point farthest in each connected domain respectively, the distance between this point and barycenter is preserved as neural length characteristic.
6., as claimed in claim 1 based on the fungal keratitis image-recognizing method of RX abnormality detection and texture analysis, it is characterized in that, the detailed process extracting the neural shapes feature of training sample in described step (2-1-3) is:
For the training sample after the process of step (2-1-2) dilation erosion, utilize Fourier transform to extract the shape facility of connected domain, be neural shape facility.
7., as claimed in claim 1 based on the fungal keratitis image-recognizing method of RX abnormality detection and texture analysis, it is characterized in that, the detailed process extracting the neural textural characteristics of training sample in described step (2-1-4) is:
The gray level image obtained after the process of step (2-1-2) dilation erosion and original image are carried out dot product, after original image dot product, obtains the gray level image of target area; Again by the method for gray level co-occurrence matrixes, neural textural characteristics is extracted to the gray level image of target area.
8. as claimed in claim 1 based on the fungal keratitis image-recognizing method of RX abnormality detection and texture analysis, it is characterized in that, in described step (2-1-5) to the detailed process that the neural length characteristic extracted, neural shapes feature and neural textural characteristics carry out Fusion Features be: the neural length characteristic value, neural shapes feature and the neural textural characteristics that extract are arranged in a row vector, as the proper vector of nerve.
9., as claimed in claim 1 based on the fungal keratitis image-recognizing method of RX abnormality detection and texture analysis, it is characterized in that, in described step (3):
Euclidean distance between Neuronal Characteristics after Neuronal Characteristics after test sample book merges and training sample merge is less, shows that the Neuronal Characteristics of the Neuronal Characteristics of test sample book and training sample is more close, belongs to same class Neuronal Characteristics;
Euclidean distance between mycelia feature after mycelia feature after test sample book merges and training sample merge is less, shows that the mycelia feature of the mycelia feature of test sample book and training sample is more close, belongs to same class Neuronal Characteristics.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510278952.4A CN104850861B (en) | 2015-05-27 | 2015-05-27 | Based on the fungal keratitis image-recognizing method of RX abnormality detection and texture analysis |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510278952.4A CN104850861B (en) | 2015-05-27 | 2015-05-27 | Based on the fungal keratitis image-recognizing method of RX abnormality detection and texture analysis |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104850861A CN104850861A (en) | 2015-08-19 |
| CN104850861B true CN104850861B (en) | 2016-04-13 |
Family
ID=53850496
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510278952.4A Active CN104850861B (en) | 2015-05-27 | 2015-05-27 | Based on the fungal keratitis image-recognizing method of RX abnormality detection and texture analysis |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104850861B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105809188B (en) * | 2016-02-26 | 2017-08-08 | 山东大学 | A kind of fungal keratitis image-recognizing method based on AMBP innovatory algorithms |
| CN108629378A (en) * | 2018-05-10 | 2018-10-09 | 上海鹰瞳医疗科技有限公司 | Image-recognizing method and equipment |
| CN108956618B (en) * | 2018-06-25 | 2020-08-11 | 衢州学院 | Tapered roller bearing surface detection method based on frequency domain vision |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101669828A (en) * | 2009-09-24 | 2010-03-17 | 复旦大学 | System for detecting pulmonary malignant tumour and benign protuberance based on PET/CT image texture characteristics |
| CN102722735A (en) * | 2012-05-24 | 2012-10-10 | 西南交通大学 | Endoscopic image lesion detection method based on fusion of global and local features |
| CN104143087A (en) * | 2014-07-24 | 2014-11-12 | 苏州大学 | Contusive retina internal segment and external segment deficiency detecting method based on SD-OCT |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR100852629B1 (en) * | 2006-06-02 | 2008-08-18 | 연세대학교 산학협력단 | Iris recognizing system and method using multifocus image sequence |
-
2015
- 2015-05-27 CN CN201510278952.4A patent/CN104850861B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101669828A (en) * | 2009-09-24 | 2010-03-17 | 复旦大学 | System for detecting pulmonary malignant tumour and benign protuberance based on PET/CT image texture characteristics |
| CN102722735A (en) * | 2012-05-24 | 2012-10-10 | 西南交通大学 | Endoscopic image lesion detection method based on fusion of global and local features |
| CN104143087A (en) * | 2014-07-24 | 2014-11-12 | 苏州大学 | Contusive retina internal segment and external segment deficiency detecting method based on SD-OCT |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104850861A (en) | 2015-08-19 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110084259B (en) | Facial paralysis grading comprehensive evaluation system combining facial texture and optical flow characteristics | |
| Crihalmeanu et al. | Enhancement and registration schemes for matching conjunctival vasculature | |
| CN110276356A (en) | Eye fundus image aneurysms recognition methods based on R-CNN | |
| CN105809188B (en) | A kind of fungal keratitis image-recognizing method based on AMBP innovatory algorithms | |
| Qureshi et al. | Detection of glaucoma based on cup-to-disc ratio using fundus images | |
| Deligiannidis et al. | Emerging trends in image processing, computer vision and pattern recognition | |
| CN104850861B (en) | Based on the fungal keratitis image-recognizing method of RX abnormality detection and texture analysis | |
| Zhu et al. | Digital image processing for ophthalmology: Detection of the optic nerve head | |
| Junjun et al. | Diabetic retinopathy detection based on deep convolutional neural networks for localization of discriminative regions | |
| Wan et al. | Retinal image enhancement using cycle-constraint adversarial network | |
| CN108596237A (en) | A kind of endoscopic polyp of colon sorting technique of LCI laser based on color and blood vessel | |
| CN112200065B (en) | Micro-expression classification method based on action amplification and self-adaptive attention area selection | |
| CN112419246A (en) | Depth detection network for quantifying esophageal mucosa IPCLs blood vessel morphological distribution | |
| Chen et al. | Pseudo-labeling for small lesion detection on diabetic retinopathy images | |
| Rahul et al. | Novel approach for detection of early diabetic retinopathy | |
| Oloumi et al. | Digital image processing for ophthalmology: Detection and modeling of retinal vascular architecture | |
| Boonchieng et al. | Edge-detection and segmentation methods for two-dimensional echocardiograms | |
| Ardizzone et al. | Blood vessels and feature points detection on retinal images | |
| Yue et al. | Deep Pyramid Network for Low-light Endoscopic Image Enhancement | |
| Liang et al. | Location of optic disk in the fundus image based on visual attention | |
| Zulfahmi et al. | Techniques for exudate detection for diabetic retinopathy | |
| CN109273081B (en) | Automatic identification system for fundus scanning | |
| Khattar et al. | Investigation of Denoising Techniques for Removal of Hair and Noise from Dermoscopic Images | |
| Giancardo et al. | Bright retinal lesions detection using color fundus images containing reflective features | |
| Ghorab et al. | Computer-Based Detection of Glaucoma Using Fundus Image Processing |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| EXSB | Decision made by sipo to initiate substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| CB03 | Change of inventor or designer information |
Inventor after: Liu Zhi Inventor after: Liu Hongtong Inventor after: Zhang Minggao Inventor after: Zhang Haixia Inventor after: Wu Xuelian Inventor after: Tao Yuan Inventor before: Liu Zhi Inventor before: Liu Hongtong Inventor before: Zhang Minggao Inventor before: Zhang Haixia Inventor before: Wu Xuelian Inventor before: Tao Yuan |
|
| COR | Change of bibliographic data | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant |