TW200821957A - Method of image object classification and identification - Google Patents

Abstract

A method of image object classification and identification, includes the following steps: (a) plotting a contour wave according to a contour curve of the image object; (b) normalizing the contour wave for adjusting the contour wave to have a normal size; (c) performing a wave transformation from space-domain to frequency-domain; (d) comparing the rotated and normalized contour wave with a wave database to search for a reference contour wave from the wave database; and (e) determining the image object to be a reference object corresponding to the reference contour wave.

Description

200821957rW3286pA IX. Description of the invention: [Technical field to which the invention pertains] The present invention relates to - Fudan 彡# + an image object classification and identification ^ & method> and especially related to [prior art] capable ^^ At present, the basic function of the surveillance video recording system is currently guaranteed, and the video recording function is currently activated or an alarm is issued. Despite the frequent misjudgment of motion detection, such as the lens system, the environment is caused by the sound of the wind blowing the leaves, or the light changes, etc. The L/, 'interference' will misidentify that the object is moving to start recording or • = ', ϊ (4) Funding _ waste and false alarm. - Steps to maintain monitoring and recording materials, combined with image processing technology, classification and identification of what kind of objects/Γ, identify that the moving object is a • eye, baggage, etc. to be monitored) Think of an alarm and greatly reduce the chance of false alarms. [Summary of the Invention] The category and the origin of the 0 (four) purpose of the difference in the cooked-Qing image object. When the application is in the fast-acquisition and accurate object classification and identification alarms, the system will reduce the mobile _ error and effectively and surely reduce the image data storage. According to the object of the present invention, an image object identification is proposed.

200821957rW3286PA .=Γα;1): One of the objects of the object is a contoured cap ^ Ageing ^ f (nG〇lalize) object wheel · ^, to scale the contour of the object to a fixed size; (c) Wave contour wave pattern Converting from the spatial domain to the frequency domain; (4) analysing the transformed contour waveform of the object and a waveform database, and finding the contour contour of the reference object from the wave pattern = ::; and (4) judging the image object One of the contour waveforms of the multi-test object corresponds to the reference object. According to the purpose of the present invention, it is further proposed that the image comprises the following steps: (4) according to the reference object: the reference object rim wave pattern, the normalized reference object edge scales the reference object contour waveform to a fixed size / In order to change the reference object contour waveform from the idling C = skin type and to convert the reference object = the above objects, features, and advantages of the present invention will be more clearly as follows - preferred implementation For example, please refer to FIG. 1 '(4) according to the present invention - 隹 倾 倾 倾 倾 倾 倾 倾 倾 倾 倾 倾 倾 倾 倾 倾 倾 倾 倾 倾 倾 倾 倾 倾 倾 倾 倾 倾 倾 倾 倾 倾 倾 倾 倾 倾 倾 倾 倾 倾 倾 倾 倾 倾Type::=According to the best way to draw the image, such as the 5A to 5C--the contour of the object is the same as the figure 8 to the % figure. As shown in Figure 5A, the image 6

200821957 W3286PA The outline curve of the object is ancient *-out and back wire" ^(xi,Yi). First, calculate the centroid coordinate (Xc, Yc) of the ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^士士! f, for example, 1 U at any point on the contour curve. Starting point in the Lfx axis direction (χο, γο) is when the scanned object is rotated, from the starting point i = Υ axis direction The first point to the point. The contour curve of β'” / σ is calculated to calculate the distance between each point and the center of gravity. _ # * # Each point coordinate (Xi, Yi) and the center of gravity coordinate (X. (8)

= can be converted into a wave', so it can be based on 1 object-object contour wave pattern: the 5C =, ,, 曰 = the coordinate graph of the contour waveform of the object's its horizontal point, and its ordinate is the distance. Ding Yuchu collar.,,. Occupy (4) 12 〇 will be the 5th (: (four) beginning of the contour contour of the object: 匕 (n:: llze), to scale the original object contour wave shape, and become the first A normalized contour wave pattern of a 5D image. Normalization == The size of the image object - that is, the size of the spring case is enlarged or reduced to a fixed value to exclude the size of the image object. The 130 series further transforms the normalization of the 5D map into a wave pattern of the corridor wave type, so that the normalized object wheel pattern is converted from the waveform of the domain to the waveform of the frequency domain, and becomes the waveform of the 5e. A normalized and converted contour waveform of the object. Waveform conversion can be performed by Fourier transform, Z transform, cosine transform, or wavelet transform, etc. The contour contour of the waveform transform can eliminate the rotation or mirroring of the image object. The opposite situation 7

200821957rW3286PA • Before the waveform comparison, the frequency fraction in the contour waveform of the object can be filtered first in step 140. Use a low-pass filter to filter out the high-frequency portion of the contour waveform of the object, using only the low-frequency portion. In this way, the high-frequency vibration cover in the contour waveform caused by the extraordinary contour of the object can be removed, and the contour waveform of the object is smoother. Next, in step 15〇, the contour contour and the waveform database of the object are converted and compared with the contour contour and the reference contour waveform. The wave type database has several parameters = round waves, which can be used for the wheel pattern of people, cars, luggage and other items. • Refer to Figure 2 for the second time, which shows the flow of the image object identification method in Figure 1. First, in step 251, the object contour waveform is selected from the waveform data, and in step 252, the selected shape=object wheel waveform is matched with the normalized and converted object, Acquired - the wave pattern is poor. Next, in step 253, the value of the wave: the pair of values or the sum of the least squares is calculated, and is defined as an alignment = i whether the value of the pair is the smallest and small - ^ data is used by the vertical step, if not, then back Go to step 25, from the wave type library, select another _ reference object contour type re-entry, wave type, so you can borrow _』:==:= pieces. The method can directly enter the last cow image object, which is the reference object with the smallest comparison value; the image object should be referenced in the gallery 6^. The method is still applicable to: the contour wave type corresponding to the auxiliary identification image object is 255 and step 256 above, and the auxiliary image is corresponding to the reference object. Step 255 is judged

200821957W3286PA • Whether the height/width ratio of the image object is within a certain range, and step 256 determines whether the color histogram of the image object conforms to the color histogram of the corresponding reference object. In addition, the method for establishing the above-mentioned waveform database is to adopt an image material classification method. Referring to FIG. 3, a flow chart of a method for classifying an image object according to a preferred embodiment of the present invention is shown. The image object identification method begins with the step _take-reference object, and in step 3iq, according to a reference object-reference contour curve (4) - reference object contour waveform. The preferred drawing method of the object f-type is also the same as the centroid coordinate (Xc, Yc) of the reference image object in the 5A to 5C; recalculation - point coordinate Wi, Yi) and the center of gravity coordinate (Xc, Yc) The distance object is ,, the target, the distance is the ordinate, the drawing-reference is used to zoom the _ wheel _ to the ambiguity; the type change, so that the positive m 物 object rim wave pattern is transferred to the - frequency domain mode : The reference object contour waveform of the waveform transformation of the spatial domain. After the wave is normalized and converted, the z-conversion, the transition to the turn-beat conversion, and the converted reference object contour waveform are stored in the wave material 1 (4). The iron-like-reference == this image object classification method is classified and then

200821957TW3286PA . Whether there is a moving object in the middle and the ======, it is better to ===== to the image screen into the background. Then, in the image storage area of the object, the moving object can be used as the image object 2 = the image object can also be used as the image object. The image object classification disclosed in the above embodiment of the present invention and (4) the large image of the image object The classification and identification of the wheel_wire; ^ The normalization of the profile to a fixed value can eliminate the situation where the size of the image is different due to the image 掏=head. In addition, the wide wheel: ίί turns to the 1 member rate domain mode 'can eliminate the image capture mirror: the resulting image object rotation and the exclusion of the trick === the opposite of the mirror shot. Therefore, through the method of fascinating in February, you can get fast and accurate classification and identification of objects and identify the results. In addition, when the present invention is applied to an image monitoring system, the π hunter quickly identifies the object of the detected moving object by recognizing the contour waveform of the moving object, and only emits a singer for the target that is actually desired to be monitored. Reporting, and greatly reduce the chance of false alarms, and can effectively and surely reduce the amount of data storage. ~mound

200821957W3286PA - In summary, although the invention has been described above in terms of a preferred embodiment, it is not intended to limit the invention. It will be apparent to those skilled in the art that various changes and modifications can be made without departing from the spirit and scope of the invention. Therefore, the scope of the invention is defined by the scope of the appended claims.

11

200821957rw3286PA - [Simplified Schematic Description] FIG. 1 is a flow chart showing an image object identification method according to a preferred embodiment of the present invention. Fig. 2 is a flow chart showing the comparison of the data bases in the image object identification method of Fig. 1. FIG. 3 is a flow chart showing a method for classifying an image object according to a preferred embodiment of the present invention. FIG. 4 is a flow chart showing image capture in the image object identification method of FIG. 1 and the image object classification method of FIG. 5A to 5E are schematic views showing the contour waveform of the formed object. [Main component symbol description] (none) 12

Claims (1)

200821957rW3286PA X. Patent application scope ··.. An image object identification method, including: type; () according to the & image object one wheel curve drawing - object wheel wave object object contour wave pattern, to zoom to The skin type conversion 'makes the contour waveform of the object from the spatial domain to the skewed normalization and the transformed contour wave pattern and the one wave type to search for a reference object contour wave pattern with the waveform type body; The husband goes to ^) to judge that the image object is one of the contour waveforms of the reference object corresponding to the Macquay object. 2. The method for identifying an image object according to claim 1, wherein the step (a) comprises: calculating a center of gravity coordinate of the image object, wherein the center of gravity coordinate is a total average of coordinates of each point on the contour curve. ; and, ★ Calculate the distance between the coordinates of each point on the contour curve and the center of gravity coordinates, and draw the contour waveform of the object accordingly. 3. The image object recognition method according to claim 1, wherein before the step (d), the method further comprises: (f) establishing the waveform database. 4. The image object identification method according to item 3 of the patent application scope, wherein the step (f) comprises: ^ ΓΨ 3286 ΡΑ 200821957 (Π) according to a reference object object contour waveform; reference contour curve edge reference (10) normalization The reference object contour waveform, the contour waveform to the fixed size; the reference object (f3) is subjected to the chopping type conversion, so that the domain is converted to the frequency domain; and (4) the waveform is spatially converted and converted The reference object rim wave type storage method, (4) 4 (four) shape object identification side calculation of the reference image object one of the reference center of gravity coordinate system for each of the reference rim curve to = average; and the point of loss pure Calculate the number of each reference point coordinate and the reference center of gravity coordinate on the reference ship's curve and calculate the surface wave shape of the ginseng piece according to the money. Method of changing 6 ^ The image object recognition method described in item 4 of the patent application section wherein the waveform conversion system of the V (1)) adopts Fourier transform, Z-turn cosine transform, or wavelet transform. L. The image object recognition method as described in the scope of the patent application, wherein before the step (fl), the method further comprises: storing the image surface without the moving object as a background image; capturing an image of the image monitoring system昼面; and - subtracting the extracted surface from the surface of the scene to separate the reference object. 200821957rw3286PA " 8. The image object identification method according to claim 1, wherein the step (d) comprises: determining a tN-hai reference object wheel-wave type and one of the normalized and converted object contour waveforms Whether the comparison value is less than a threshold value. 9. The method for identifying an image object according to item 8 of the patent application, wherein the step (d) comprises: ... separating a plurality of the contour waveforms of the reference object from the 4-wave database, and normalizing and converting the image The contour waveform of the object is subtracted to obtain a plurality of waveform differences; and the absolute value or the sum of the least squares of the waveform differences is defined as a plurality of the comparison values; and ^ is selected A minimum alignment value for one of the values, and the minimum alignment value is less than a threshold. 10. The method for identifying an image object according to item 8 of the patent application, wherein the step (d) further comprises: determining whether the aspect ratio of the image object is within a specific range, and assisting in identifying the image object system. For this corresponding reference object. The image object identification method of claim 8, wherein the step (4) further comprises: determining whether the color histogram of the image object conforms to the color histogram of the corresponding reference object to assist in identifying the image object as the image Corresponding reference object. 12. The method for identifying an image object as described in claim 1 of the scope of the patent application wherein the waveform conversion of the step (c) is Fourier transform, 2 revolution, 200821957rW3286PA, commutation, cosine transform, or wavelet transform. 13) The method for identifying an image object according to the application scope of the patent application, wherein, before the step (a), the method further comprises: storing the image surface without the moving object as a background surface; capturing a scene> An image image of the surveillance system; and the surface of the image is removed from the background surface to separate the image object. Person 14. The method of claim 1, wherein before step (4), the method further comprises: absorbing the high frequency portion of the contour waveform of the object. 15·—A method for classifying image objects, comprising: (a) drawing a reference object according to a reference contour curve % of a reference object The center of gravity coordinate is the average; and 16 200821957 W3286PA - Calculates the distance between each reference point coordinate on the reference contour and the reference centroid coordinate, and draws the reference object contour waveform accordingly. 17. The image object classification method according to claim 15, wherein the waveform conversion of the step (c) is a Fourier transform, a Z-transform, a cosine transform, or a wavelet transform. 18. The image object classification method according to claim 15, wherein before the step (a), the method further comprises: storing the image surface without moving objects into a background surface; capturing an image monitoring system The image is imaged; and the subtraction surface is subtracted from the background image to separate the reference object of the displacement. 17