Template Matching Process Research Articles

ZNCC-based template matching using bounded partial correlation

This paper describes a class of algorithms enabling efficient and exhaustive matching of a template into an image based on the Zero mean Normalized Cross-Correlation function (ZNCC). The approach consists in checking at each image position two sufficient conditions obtained at a reduced computational cost. This allows to skip rapidly most of the expensive calculations required to evaluate the ZNCC at those image points that cannot improve the best correlation score found so far. The algorithms shown in this paper generalize and extend the concept of Bounded Partial Correlation (BPC), previously devised for a template matching process based on the Normalized Cross-Correlation function (NCC).

High-Speed Data Position Detection Method Using Template Matching for Hologram Memory

Two-dimensional digital image reproduction in a hologram memory causes incorrect decoding due to image distortion. In order to correct image distortion, the position coordinates of the four corners of the reproduced image are detected by template matching. However, much position detection processing time is required in this signal processing method. In this study, after having used both parallel processing that combines the marker-dividing approach and marker detection image-dividing approach in the hardware detecting process, and the redundant calculation elimination method and brightness totaling method in the software detection process, it is clarified that an increase in the speed of 260-fold could be obtained as compared with the conventional template matching process.

Foreign Body Detection by Estimating Voted Results Using Block Image Matching

This paper proposes a method to detect an image area that is occluded by a foreign body in the template matching process. The template image is separated into block areas and the normalized correlation matching is executed by each separated block area. Next, the voting process about each block area that increments the value in the address with the maximum matching rate is executed in the two-dimensional space. The location of the target image area is obtained by the address with the maximum votes. The partial blocks without the proper voted address mean the image area occluded by a foreign body. The experimental results show that the proposed method detected the occluded area by the foreign body in the template matching process.

Fast algorithm for robust template matching with m-estimators

We propose a fast algorithm for speeding up the process of template matching that uses M-estimators for dealing with outliers. We propose a particular image hierarchy called the p-pyramid that can be exploited to generate a list of ascending lower bounds of the minimal matching errors when a nondecreasing robust error measure is adopted. Then, the set of lower bounds can be used to prune the search of the p-pyramid, and a fast algorithm is thereby developed in this paper. This fast algorithm ensures finding the global minimum of the robust template matching problem in which a nondecreasing M-estimator serves as an error measure. Experimental results demonstrate the effectiveness of our method.

Fast template matching using correlation‐based adaptive predictive search

AbstractWe have developed the Correlation‐based Adaptive Predictive Search (CAPS) as a fast search strategy for multidimensional template matching. A 2D template is analyzed, and certain characteristics are computed from its autocorrelation. The extracted information is then used to speed up the search procedure. This method provides a significant improvement in computation time while retaining the accuracy of traditional full‐search matching. We have extended CAPS to three and higher dimensions. An example of the third dimension is rotation where rotated targets can be located while again substantially reducing the computational requirements. CAPS can also be applied in multiple steps to further speed up the template matching process. Experiments were conducted to evaluate the performance of 2D, 3D, and multiple‐step CAPS algorithms. Compared to the conventional full‐search method, we achieved speedup ratios of up to 66.5 and 145 with 2D and 3D CAPS, respectively. © 2003 Wiley Periodicals, Inc. Int J Imaging Syst Technol 13, 169–178, 2003; Published online in Wiley InterScience (www.interscience.wiley.com). DOI 10.1002/ima.10055

Linearity Measurement of Welded Part Using Self-Renewed Template Image Matching

It is important to measure the linearity of a welded part to evaluate the quality of manual welding. This paper proposes a method to measure the linearity of a welded part that has an uneven surface condition by image processing. The initial template image that includes a part of a welded area is set at the left edge in the image. Next, the template image matching is repeated to the right edge continuously. In the template matching process, the template image is renewed to the searched image pattern at every matching location. The vertical differential position of the welded part is measured in the continuous template matching process and their standard deviation is calculated as the linearity. The experimental results show that the curve obtained by the proposed method expresses the global form of the welded part appropriately. The linearity measured by the proposed method correlated closely with the linearity measured by the visual sense.

Development of template matching hardware and its high‐speed processing strategy

AbstractThe template matching technique is now applied to tracking of outdoor object and position measurement, and there is increasing demand for application of template matching based on the normalized correlation coefficient, which is robust against the variation of illumination. In order to apply the technique based on the normalization technique to practical application problems, however, high‐speed calculation of the normalized correlation coefficient is required. This paper intends to improve the speed by implementing by hardware the calculation of the normalized correlation coefficient, which requires a large amount of computation. The proposed method decomposes the normalized correlation coefficient calculation into five operational components, each of which is implementated by hardware. The speed of processing is improved being combined with a parallel processing with a high degree of freedom based on the local property owned by the template matching process. For further improvement of the speed, the application of the pyramid structure is considered. Hardware that can handle the pyramid structure is realized. An evaluation was tried using an actual machine. The performance of the developed hardware is evaluated. The effectiveness of the approach is verified by applying the method to the traffic flow measurement system, as an example of an actual outdoor system. © 2001 Scripta Technica, Electron Comm Jpn Pt 3, 84(11): 1–10, 2001

Detecting human faces in color images

A method for detecting human faces in color images is described that first separates skin regions from nonskin regions and then locates faces within skin regions. A chroma chart is prepared via a training process that contains the likelihoods of different colors representing the skin. Using the chroma chart, a color image is transformed into a gray scale image in such a way that the gray value at a pixel shows the likelihood of the pixel representing the skin. An obtained gray scale image is then segmented to skin and nonskin regions, and model faces representing front- and side-view faces are used in a template-matching process to detect faces within skin regions. The false-positive and false-negative errors of the proposed face-detection method on color images of size 300×220 and containing four or fewer faces are 0.04 or less.

SHAPE-BASED RETRIEVAL: A CASE STUDY WITH TRADEMARK IMAGE DATABASES

Retrieval efficiency and accuracy are two important issues in designing a content-based database retrieval system. We propose a method for trademark image database retrieval based on object shape information that would supplement traditional text-based retrieval systems. This system achieves both the desired efficiency and accuracy using a two-stage hierarchy: in the first stage, simple and easily computable shape features are used to quickly browse through the database to generate a moderate number of plausible retrievals when a query is presented; in the second stage, the candidates from the first stage are screened using a deformable template matching process to discard spurious matches. We have tested the algorithm using hand drawn queries on a trademark database containing 1100 images. Each retrieval takes a reasonable amount of computation time (∼4–5 s on a Sun Space 20 workstation). The topmost image retrieved by the system agrees with that obtained by human subjects, but there are significant differences between the ranking of the top-10 images retrieved by our system and the ranking of those selected by the human subjects. This demonstrates the need for developing shape features that are better able to capture human perceptual similarity of shapes. An improved heuristic has been suggested for more accurate retrievals. The proposed scheme matches filled-in query images against filled-in images from the database, thus using only the gross details in the image. Experiments with database images used as query images have shown that matching on the filled-in database extracts more images within the top-20 retrievals that have similar content. We believe that developing an automatic retrieval algorithm which matches human performance is an extremely difficult and challenging task. However, considering the substantial amount of time and effort needed for a manual retrieval from a large image database, an automatic shape-based retrieval technique can significantly simplify the retrieval task.

Chinese document layout analysis based on adaptive split-and-merge and qualitative spatial reasoning

The ultimate goal of automatic document processing is to understand the semantics of a document. Towards such an end, one of the primary enabling steps has been to first reason about the layout of the document by means of page segmentation and segment spatial reasoning or labeling. This, in turn, allows for the derivation of document logical organization. This paper describes a generic document segmentation and geometric relation labeling method with applications to Chinese document analysis. Unlike the previous document segmentation methods where text spacing, border lines, and/or a priori layout models based on template matching processing are performed, the present method begins with a hierarchy of partitioned image layers where inhomogeneous higher-level regions are recursively partitioned into lower-level rectangular subregions and at the same time lower-level smaller homogeneous regions are merged into larger homogeneous regions. Furthermore, the derived segment data structure readily enables efficient search for geometric relationships between identified document segments.

Inductive neural logic network and the SCM algorithm

Neural Logic Network (NLN) is a class of neural network models that performs both pattern processing and logical inferencing. This article presents a procedure for NLN to learn multi-dimensional mapping of both binary and analog data. The procedure, known as the Supervised Clustering and Matching (SCM) algorithm, provides a means of inferring inductive knowledge from databases. In contrast to gradient descent error correction methods, pattern mapping is learned by an inductive NLN using fast and incremental clustering of input and output patterns. In addition, learning/encoding only takes place when both the input and output match criteria are satisfied in a template matching process. To handle sparse and/or noisy data sets, we also present a weighted voting scheme whereby distributed cluster activities combine to produce a final output. The performance of the SCM algorithm, compared with alternative systems, is illustrated on three benchmark problems: (1) mushroom classification, (2) sonar return signal recognition, and (3) sunspot time series prediction.

B212 DPマッチング法による画像流速測定について

New PIV measurement using a DP matching cross-correlation method is attempted on the experimental flows showing a 2-D circular cylinder wake. The merit of this matching method is the capability of admitting a certain degree of rotation and/or strain deformation of the correlation templates. In addition, since the concept of DP matching has the idea of checking the image shift parameters at the initial stage of each template matching, the computing time required for whole field measurement is shorter than the conventional cross-correlation method, despite more complicated process of template matching. In all of the present test cases, the resultant vector maps are more or less improved from the usual crosscorrelation method.

Improved recovery of high frequency P wave energy by selective P wave averaging.

Most signal averaging processes used for cardiac signals align successive waveforms using a template matching process. In addition to achieving accurate temporal alignment of the signals, this operation must also ensure that the signal average comprises signals of the same morphology. For P wave signal averaging, systems designed for QRS complex averaging are often used, with the template acquisition window shifted to include the P wave. Theoretically, with this technique, variations in P wave morphology could reduce the high frequency content of the signal. We tested this hypothesis by comparing the performance of a selective P wave averaging system with a conventional system, based on template matching by cross-correlation over a fixed acquisition window, on identical P wave recordings from 15 subjects. The selective system identifies variations in P wave morphology and generates up to five candidate templates for averaging. Subsequently, the most frequently matched template over a 100-beat sample is used for averaging. Only P waves with the same morphology as this template are averaged. Selective averaging mainly increased the measured high frequency P wave energy, without affecting P wave duration after high pass filtering at 40 Hz, (Duration: 157(4)ms selective vs 155(4) nonselective. Energy 80-150 Hz: 1.77(0.28), microV2.s selective vs 1.61(0.3) microV2.s non selective, P < 0.01). These observations confirm that nonselective P wave averaging can reduce apparent P wave energy, especially at high frequency. Before meaningful studies of the value of frequency domain analysis of the P wave can be performed, it is important that signal averaging systems of sufficient fidelity are utilized.

Speech recognition training method

A speech recognition method and apparatus employ a speech processing circuitry for repetitively deriving from a speech input, at a frame repetition rate, a plurality of acoustic parameters. The acoustic parameters represent the speech input signal for a frame time. A plurality of template matching and cost processing circuitries are connected to a system bus, along with the speech processing circuitry, for determining, or identifying, the speech units in the input speech, by comparing the acoustic parameters with stored template patterns. The apparatus can be expanded by adding more template matching and cost processing circuitry to the bus thereby increasing the speech recognition capacity of the apparatus. Template pattern generation is advantageously aided by using a "joker" word to specify the time boundaries of utterances spoken in isolation, by finding the beginning and ending of an utterance surrounded by silence.

Voice-activated word processor with automatic learning for dynamic optimization of syllable-templates.

A voice-activated word processor has been realized which provides text input by articulated, phrase-by-phrase Japanese speech, using an automatic learning algorithm to improve syllable recognition accuracy. A speaker-dependent, continuous-phrase speech recognizer incorporated into the word processor is capable of real-time recognition using 111 monosyllables as the basic recognition units. The recognizer is trained by new user to construct 590 reference syllable templates based on uttering a set of words, necessary for syllable template-matching process. Syllable templates are continually updated during use of the word processor through automatic learning. The automatic learning algorithm replaces low-accuracy syllable templates with new patterns extracted from the input speech. The replacement is sensitive to syllable context and is carried out based on recent and longer term history of the recognition accuracy of each syllable. Such learning information is derived from comparison between syllable recognition results and a user-confirmed character string of input phrase as correct, and is accumulated on a phrase-by-phrase basis. The automatic learning algorithm was tested in experiments using Japanese sentences read with pauses between phrases at approximately 4 to 5 syllables per second. The results, using eight speakers, show average syllable recognition accuracy of 82.5% with automatic learning, compared to 71.0% achieved without the learning. Further, the recognition accuracy is increased to 86.5% when the maximum number of syllable templates is increased to 2, 048; both template replacement and template addition are carried out until the maximum number of templates is reached.

Computer-based determination of size and shape in living cells.

Measurement of cell volume in living epithelial cells has become an important technique in studies of membrane transport processes that function in cell volume regulation. Planimetry of video images of optical sections enables the measurement of the cross sectional area of each section. Cell volume is calculated from the measured area of each section and the known focus displacements. In the past the measurement of cross section area has been done by manual positioning of a cursor superimposed on the video image. Each experiment generates approximately 200 images in which two or more cells may be analysed. We have developed a computer-based method that uses one image as a template, and allows automated area determination of successive images by template matching and digital image processing. This new method is comparable to the older method in speed and accuracy, but requires much less effort from the experimenter.

The effect of uncertainty on the detection of frequency modulation at low modulation rates.

Frequency modulation detection performance is compared in two conditions of stimulus variability. The data show that when the modulation rate is randomized within an experimental block, performance is somewhat poorer than when the modulation rate is fixed. The results are interpreted within the framework of a template matching model for modulation detection. They suggest that a template is acquired in part from experimental context and in part from a multistage template matching process.

Storage mode of geometrical dimensions in a two-stimulus matching paradigm.

In a two-stimulus matching paradigm requiring “same” and “different” responses to simple geometric dimensions, the successive stimulus presentation mode with several retention intervals was compared to a simultaneous stimulus presentation mode. For the simultaneous condition, the data suggested a wholistic template matching process, but in none of the successive conditions was template matching indicated. The retention interval of the successive condition allows the subject to analyze out the target dimension from the first two-dimensional stimulus. The ability to anticipate the value of the target dimension in the second stimulus could also influence the representation of the first stimulus. The findings shed some light upon the perceptual interactions between dimensions that constitute the stimulus. In particular, an outside-to-inside order of iconic scanning appeared to be supported by the data.

Experiments in iterative enhancement of linear features

Lines and curves in an image are detected locally by a template-matching process which determines the “line-ness” value of the image at each point, in a set of orientations. The output of the detection process is the strongest of these values at each point, and the orientation that gave rise to this value. The results of this approach tend to be noisy, but their noisiness can be reduced by examining, for each point, the values at nearby points, in the direction defined by the preferred orientation, and increasing the point's value if the nearby points have high values and similar orientations. Iteration of this reinforcement process leads to further noise reduction. Several variations on this scheme are presented. The preferred orientations can also be “sharpened” by examining the orientation at nearby points (in the preferred direction) and biasing it toward their average. Experimental results using these methods are obtained for a portion ef a LANDSAT image containing many linear features.