طراحی فیلتر, فیلتر تجارت کردن کامپوزیت با مناطق پشتیبانی جهت به دست آوردن شناخت الگوی ثابت با تصاویر ضد فوکوس
|کد مقاله||سال انتشار||مقاله انگلیسی||ترجمه فارسی||تعداد کلمات|
|22351||2003||13 صفحه PDF||سفارش دهید||محاسبه نشده|
Publisher : Elsevier - Science Direct (الزویر - ساینس دایرکت)
Journal : Optics and Lasers in Engineering, Volume 40, Issues 1–2, July–August 2003, Pages 67–79
The discrimination capacity (DC) measures the ability of the filter in a pattern recognition problem to discriminate the target against other objects in the input scene. If the input scene is degraded by a defect of focus, then the DC is degraded and the pattern recognition process is worse. In this paper, we present a methodology based in the selection of ring frequency bands and in the design of the trade-off filters taking into account these frequencies to obtain several information channels. The information of all the channels is fused by means of the addition of all the channels and the geometric mean of them. Also individual channel analysis is shown. The influence on the DC and SNR of the added white noise in the input image is presented.
In pattern recognition problem, the filter design has been a very important development area. The success of the recognition process strongly depends on the used filters, so many filters have been proposed each one optimizing some quality criteria or solving a determinate task. The start point of some of these tasks is the quality of the input image to be processed. So, depending on the kind of the degradation, the solution to obtain better results will be different. For instance, if the input image presents rotation in plane, a filter based on circular harmonic decomposition permits to obtain invariant results . There are problems that cannot be solved by the use of only one filter being necessary the use of several filters which give multiple information channels each one with their own information. In the literature several architectures have been proposed to optically implement multichannel filtering. Yu et al.  proposed a technique to perform multiple correlation that uses a scanning grating displayed on a spatial light modulator (SLM) where the orientation of the grating is sequentially varied. Sheng et al.  and Mendlovic et al.  proposed a multichannel optical processors for implementation of the wavelet transform. The replicas produced by the pixelated structure of the spatial light modulators have been used to implement a multichannel pattern recognition  and Gabor-wavelet transform . In  the design of a gray-level computer generated hologram filter for multiple-object correlation is provided. The impulse response (IR) of this filter is the combination of the IR of several filters spatially disjoined which permits the implementation of a multichannel process. Some times the input image appears distorted or degraded. Then especial filters are needed for pattern recognition. For instance, Chang et al.  have proposed a new method for pattern recognition that is invariant under changes of position, orientation, intensity and scale. It is based on the centroids of objects. A common degradation that appears in the input scene is the defect of focus. To obtain a less sensitive process to the degradation introduced in the input scene by a defect of focus of the imaging system we have used  a Laplacian pyramidal decomposition over the input scene. A sub-band image set is obtained, each sub-band image is used to perform a correlation process with a POF filter. In this way a multichannel process is carried out. By means of a non-linear combination of the information provided by all channels invariance against the defocus can be obtained. Due to the restriction between the central frequency band and the width of the Laplacian filter, better results are obtained when annular preprocessing filter  is used. In this paper, we propose the design of a trade-off SDF filter in annular regions of support to obtain an effective and robust process invariant to defocus degradation in the input scene. We will study the behavior of the method when only a filter or the combination of all the channels is used. In Section 2, we give an overview of the out of focus effect on the correlation process, and why regions of support of annular shape are selected. In Section 3, the filter design is described. Finally, in Section 4 several experiments are carried out to study the behavior of the proposed method. Then in Section 5 we present the conclusions.
نتیجه گیری انگلیسی
We have presented a methodology to obtain an improvement in the discrimination capability in the pattern recognition process when the input image suffers from a defect of focus degradation. We have proposed the use of a decomposition method with ring frequency bands and the design of trade-off filter taking into account these bands. We have analyzed the behavior of the method when only one channel or when the combination of all the channels is used. Two methods of combination are used: the addition and the geometric mean. In the case of only one channel we obtain that the DC increases when the central frequency increases of the used band for all the studied defocus W2,0. But the DC value can be low if that channel contains some zero of the OTF of the optical acquisition system. When a combinations of all the channels is used the results are much better. The correlations with the objects to be rejected are very low, and the correlations with the object to be recognized have similar values. These results are maintained for all the defocus parameter W2,0 studied. Also we have analyzed the robustness to noise of the method. In the case of single channels, lower frequency channels are more resistant to noise than higher frequency channels. In the case of addition and geometric mean combinations the process is more robust to the noise that when we use only one information channel. This behavior is maintained for all the defocus parameter W2,0 studied. The DC values to reject an object of the input image is deteriorates when W2,0 and variance increases. The addition of the channels is more robust against noise than the geometric mean. On the contrary, if the noise corrupting the scene is very low, the DC is more constant against defocus when the geometric mean of the channels is used.