The automatic design of a 3D tooth model plays a crucial role in dental digitization. However, current approaches face challenges in compositional 3D tooth generation because both the layouts and shapes of missing teeth need to be optimized. In addition, collision conflicts are often omitted in 3D Gaussian-based compositional 3D generation, where objects may intersect with each other due to the absence of explicit geometric information on the object surfaces. Motivated by graph generation through diffusion models and collision detection using 3D Gaussians, we propose an approach named DM-CFO for compositional tooth generation, where the layout of missing teeth is progressively restored during the denoising phase under both text and graph constraints. Then, the Gaussian parameters of each layout-guided tooth and the entire jaw are alternately updated using score distillation sampling (SDS). Furthermore, a regularization term based on the distances between the 3D Gaussians of neighboring teeth and the anchor tooth is introduced to penalize tooth intersections. Experimental results on three tooth-design datasets demonstrate that our approach significantly improves the multiview consistency and realism of the generated teeth compared with existing methods.

Gender-based violence affects millions of women worldwide, transcending cultural, economic, and social boundaries.According to an ONU survey, home is the most dangerous place.In a large number of cases, even with the possibility of local telephone numbers for help, at-risk women cannot make a call simply because the aggressor is close to them.Considering the accessibility of communication for the deaf, there are tools to teach American Sign Language (ALS), which is also used in countries other than the United States.The Signal for Help was created by the Canadian Women's Foundation, allowing women to silently send an SOS.The idea presented in this work combines a fake app to teach sign language with a way to use this signal without the possible note of any neighbor aggressor.This is a silent and effective request to help at-risk women.After implementing the first version, it was tested by users who suggested significant improvements, and a second version was developed.This version was also sent to users for experimentation, and an update to the current version is presented here.Thus, there are useful functionalities in the application that fulfill the needs of daily use, such as access by voice activation, login with facial recognition, interface with contrasting colors, and the possibility of setting the language of the screen texts and menu words, among others.The geographical location of the woman sending the signal helps in her location, allows quick contact with the protection service in the region, and helps in identifying areas of greater risk (geographic tracking map) for the regional public security authorities.

Breast cancer remains a significant global health challenge, where early detection critically improves patient outcomes.Thermography presents a promising non-invasive imaging modality for screening.This paper details a work-in-progress on a computer vision application utilizing thermography images for breast cancer prediction.We focus on the implementation of a scalable, reproducible, and accessible machine learning pipeline built with Apache Airflow, Docker Compose, FastAPI, Gradio, MLflow, and PyTorch Lightning.The objective is to bridge the gap between research prototypes and practical deployment, facilitating easier testing and use, particularly in resource-constrained settings.We present the system architecture, initial methodological considerations addressing data leakage found in related work, and preliminary results, seeking feedback for further development.

This paper presents a novel approach for 3D breast reconstruction using infrared (IR) images, leveraging Gaussian Splatting (3D-GS) techniques to improve diagnostic accuracy in breast cancer detection.The proposed method adopts the 3D-GS based method pixelSplat to generate detailed 3D models of breast tissue, aiding in the precise detection of abnormalities.The application of pixelSplat allows for real-time, memory-efficient rendering, improving the resolution of 3D models while overcoming common challenges in medical imaging, such as the handling of sparse data and the absence of camera pose information.By utilizing infrared thermography, this work addresses the challenge of enhancing early detection and treatment planning, ultimately contributing to more effective and personalized care within the public health system.This innovative approach holds significant potential for advancing breast cancer detection and treatment strategies in clinical practice.

One of the most prevalent types of cancer worldwide is breast cancer, a condition that tends to change the thermal pattern of the breasts. Thermographic images, a functional examination that considers temperature variation, emerge as an alternative for this disease since they consider body temperature variation to investigate anomalies. This examination has been widely investigated in studies for screening or diagnosing breast cancer. However, a limited number of studies investigate this examination to track the progress of cancer and assess tumor response to treatment. This study works in this context, exploring thermography during neoadjuvant treatment. In the proposed methodology, we first preprocess the thermal data and use the k-means unsupervised learning algorithm to identify the hottest regions. Subsequently, we build time series based on statistical measures and homogeneity measures among thermal captures to evaluate the patients’ tumor evolution. The results show that this approach indicates the treatment evolution correctly in at least 79% of the cases when observing only the statistical measures and 95% of the cases when combining the statistical and homogeneity measures on the patient evaluation process.

This paper presents an approach for detection and quantification, with low latency, of the flow of leakage bubbles, in sub-surfaces, making use of video recorded by remote underwater vehicle using only image analysis and under the premise of no overlapping bubbles. Implementation details are presented allowing its trial and reproduction. Results are confronted with videos acquired in a laboratory under controlled conditions and in real operational situation from literature, showing great efficiency in terms of processing time and all other important aspects for pipeline inspections, considering environment and safety in the oil industry.

Ultrasonography (US) has demonstrated many advantages in the detection, characterization, and monitoring of different diseases. Through high frequency probes, it is possible to visualize and characterize the anatomical layers, such as tissues, which can constitute a very helpful supporting tool in several procedures, e.g., surgeries. However, the visual identification of tissues in this type of image is still a challenge for some professionals. In this work, we evaluate deep learning (DL) segmentation algorithms for the tissue segmentation task in US. Moreover, we briefly assess whether their performance can be improved by including a crowdsourcing step. In order to perform the segmentation task, different segmentation models are trained, as posteriorly, the crowdsourcing step is included. The proposed approach is composed of the following steps: 1 - automatic segmentation using a deep learning algorithm; 2 - crowd evaluation and correction of the results. In order to perform step 1, different segmentation models are trained. The second step includes a visual interface where users can: a) validate the quality of the automatic segmentation; or b) correct the segmentation whether the DL result is inconsistent. All users are scored, denoting the quality of their annotations, considering the manual annotations provided by an expert for a small group of images. In order to evaluate the method and compare it to a DL algorithms alone, a total of 100 US images were used. Our experiments show that the inclusion of crowdsourcing significantly improved the performance of the tissue segmentation task compared to using the DL models alone. The performance of our method demonstrated the feasibility of applying this type of solution for the considered problem of segmenting tissues in US facial images. Moreover, the results suggest that this tool can be employed as an auxiliary tool in oral procedures.

Voice is the fastest and most efficient method of communication among humans. Researchers believe that it can also be considered the most efficient means of communication between humans and machines. Voice can, in addition to providing useful information, inform us about the emotional state of the person who is speaking. For many applications, being able to identify the emotion is crucial, as it allows the application to adapt to the user. In the case of human-robot interaction, recognizing the user's emotion allows the robot to be more empathetic during interactions. In addition to other methods, such as recognition of facial expressions and recognition through body expressions, recognition of emotions through speech can be used as an additional component in identification the user's emotional state. This work proposes the recognition of emotion through speech using an approach based on image processing of the voice audio signal spectrogram. Two new features based on color histograms are proposed. One thousand six hundred audio files with phrases considering four types of emotions (angry, happy, neutral and sad) were processed and classified. These phrases were spoken by women half by a 64 years old (Subject- 64) and the rest by a 26 years old one (Subject-26). These files are a subset of the TESS (Toronto Emotional Speech Set) dataset. When processing subject-26's voice, an precision of 94.40 % and 91.90 % was achieved in detecting neutral and sad emotions, respectively. When processing subject-64's voice, an precision of 97.00 % was achieved for the angry emotion. The results obtained show the proposal great potential.

Several studies explore the use of thermography for screening or diagnosis of breast cancer. However, few investigate this examination to follow up breast cancer neoadjuvant treatment. This work explores this theme and proposes a computational methodology for monitoring breast cancer treatment response by creating time series from dynamic thermography examinations. In this approach, we use the unsupervised learning algorithm k-means to cluster breast temperature information and indicate the area used for creating time series. From each time series, we compute two variation measures. Our results are promising and show that these time series measures can indicate the patient's tumor evolution.

Periodontitis is an oral disease that promotes not only inflammation but also tissue d estruction, which is visible in periapical radiographs. Early diagnosis is essential to prevent the progression of this lesion. This work's main objective is to classify regions in periapical examinations according to the presence of periodontal bone destruction. This study considered 1079 interproximal regions extracted from 467 periapical radiographs. This data was annotated by experts and used to train a ResNet and an Inception model, which were after evaluated with a test set. Inception presented the best results and an impressive rate of correctness even on the small and unbalanced dataset. The final accuracy, precision, recall, specificity, and negative predictive values are 0.817, 0.762, 0.923, 0.711, and 0.902, respectively. The ROC and PR curves also demonstrate the good performance of both models. These results suggest that the evaluated CNN model can be used as a clinical decision support tool to diagnose periodontal bone destruction in periapical exams.

The BI-RADS report system is widely used by radiologists and clinicians to document relevant findings in the mammogram exam by using a 6 category final assessment. Deep learning has achieved a high level of accuracy in multi category classification of natural images. Because of that, it is of interest to address the mammography malignancy classification according to the established BI-RADS categories. In this work, we use transfer learning on NASNet Mobile and fine tuning on VGG16 and VGG19 to classify mammogram images according to the BI-RADS scale on the INbreast dataset. Our proposed methodology achieved an accuracy (ACC) of 90.9% and a macro averaged area under the receiver operating characteristic curve (AUC) of 99.0%; outperforming some of the similar works found in the literature review.

A radiological study is comprised by a set of images together with a medical report, which is generated by radiologists to describe the main characteristics of such images and the associated clinical findings. The radiological study provides relevant information about the patient's health condition which is necessary for physicians to accomplish diagnosis. However, some works demonstrated that the reports can be vague, incomplete, ambiguous or having inaccuracies, inconsistencies or errors. In this paper, we propose to use ontologies to support the elaboration of radiological reports. We propose an ontology to represent the medical knowledge about cranial computed tomography (CCT), which is among the most required radiological studies in emergency or regular treatments. Finally, we evaluate the quality of the reports generated based on this ontology.

Breast cancer is the second most common cancer in the world. Early diagnosis and treatment increase the patient's chances of healing. The temperature of cancerous tissues is generally higher than that of healthy neighbouring tissues, making thermography an option to be considered in screening strategies for this type of cancer. In this paper, we propose a computational method for breast Dynamic Infrared Thermography images analysis for screening patients with abnormalities in the breast, using supervised and unsupervised machine learning techniques. An abnormality may be a benign tumor or a malignant tumor (cancer). As performance measure, we use the area under ROC curve, sensitivity, specificity and accuracy. The best results are achieved by K-Star classifier, obtaining an accuracy equal to 98.57%. The results confirm the potential of the proposed method for screening patients with abnormalities in the breast.

Medical images are important elements for the diagnosis of diseases. Computer Aided Diagnostic has evolved in recent years along with the processing capacity of computers as well as the emergence of new computational techniques. Segmentation is a valuable approach for identifying a specific area in human body images, such as the lungs and heart. This work proposes an algorithm to segment anatomical structures using parallel 3D region growing. Experiments using different Computer Tomography scans show that the proposed approach can run 150 times faster than the typical sequential region growing algorithm while providing good results in the identification of the target region.

Vascular structures in the retina contain important information for the detection and analysis of ocular diseases, including age-related macular degeneration, diabetic retinopathy and glaucoma. Commonly used modalities in diagnosis of these diseases are fundus photography, scanning laser ophthalmoscope (SLO) and fluorescein angiography (FA). Typically, retinal vessel segmentation is carried out either manually or interactively, which makes it time consuming and prone to human errors. In this research, we propose a new multi-modal framework for vessel segmentation called ELEMENT (vEsseL sEgmentation using Machine lEarning and coNnecTivity). This framework consists of feature extraction and pixel-based classification using region growing and machine learning. The proposed features capture complementary evidence based on grey level and vessel connectivity properties. The latter information is seamlessly propagated through the pixels at the classification phase. ELEMENT reduces inconsistencies and speeds up the segmentation throughput. We analyze and compare the performance of the proposed approach against state-of-the-art vessel segmentation algorithms in three major groups of experiments, for each of the ocular modalities. Our method produced higher overall performance, with an overall accuracy of 97.40%, compared to 25 of the 26 state-of-the-art approaches, including six works based on deep learning, evaluated on the widely known DRIVE fundus image dataset. In the case of the STARE, CHASE-DB, VAMPIRE FA, IOSTAR SLO and RC-SLO datasets, the proposed framework outperformed all of the state-of-the-art methods with accuracies of 98.27%, 97.78%, 98.34%, 98.04% and 98.35%, respectively.

Breast cancer is one of the cancer types most commonly diagnosed among women worldwide. Diagnostic techniques are constantly being developed. Dynamic thermography emerges as a tool to aid in this process. Images captured under dynamic protocol were used here to obtain breast behavior on achieving thermal equilibrium. The regions of interest (ROIs) are segmented from these images and used for analysis of the temperatures during the time of the exam. Features based on statistic and clustering are used in these analyses. Time series are formed with these features using combinations of intervals constructing subsets of different cardinalities for following their evolution over time. Groups of features are classified by Support Vector Machine, using the Leave-One-Out Cross-Validation method. Achieved results for classifications on healthy or abnormal breast from a sample of 64 breasts (half healthy and half with some abnormality) are presented.

Disorder of the thyroid glands is a widespread health problem. Early detection of thyroid cancer increases the chances of effective treatment. Dynamic infrared thermal imaging (DITI) is an examination technique that has been recently studied and applied for investigation and diagnosis of different diseases. Patterns allowing differentiate regions of malignant from benignant nodules is a task of great importance in DITI. In this work, two techniques of analysis of thyroid nodules with infrared thermography are investigated: the use of a single thermogram and the use of temperature series. The images used are available for public use by other researchers, and the used techniques are completely described as well as the achieved results. No other works using infrared images until now have considered DITI examination for thyroid nodules investigation. Moreover, it is the first work to release the used images for public use and possible future comparison.

Most studies analyzing medical images at some stage require the demarcation of boundaries of biological structures. This process is called segmentation. In some contexts, current techniques present satisfactory results, but in others, like breast segmentation in thermographies, it remains an open problem. Several studies have investigated the use of automated solutions for this problem. However, the automatic process does not always present a satisfactory result, requiring the active involvement of a specialist for validating it and re-segmenting images when necessary. As such task can be expensive and take too long to be completed, this scenario drives the exploration of alternative approaches for the segmentation process. Hence, in this work we propose an alternative that combines traditional techniques of image processing with techniques of collective intelligence, which is based on the wisdom of crowds to solve problems in a faster and less expensive way. We present SegMedBC, a prototype in which the methods previously mentioned are applied to improve the segmentation process. Furthermore, an experimental study is carried out to validate the involvement of lay users in this activity.

Thyroid anomalies have high prevalence and their early identification is crucial for a more effective treatment. Thermograms can be used in this process, since nodules tend to be more vascularized, resulting in a different behaviour delated to temperatures on the skin surface over them. This work presents a methodology for determining thyroid nodules. We evaluate features that would allow to segment possibly nodular regions in thermographs. Convolutional neural networks (CNN) are used to classify these regions, identifying which ones refer to nodules. The good results of CNN in the classification (with more than 92% of accuracy), show that the viability of the proposed methodology depends on the success of the segmentation.

Several computational activities, such as image segmentation and classification, use texture information. This information, also called descriptors, is typically calculated through the Gray Level Occurrence Matrix (GLCM), which has a quadratic cost. In this work, we discuss an alternative to this method using Sum and Difference Histograms (SDH), which has linear cost. A set of nine equations that already exist in the literature, but presenting high values of difference with GLCM, were investigated and adapted in this study. The differences (which in some cases were very high) were eliminated in most of the descriptors. The few remaining differences present values very close to zero, with a ratio of 10-1. In addition, six other equations are proposed and discussed in this study. To validate these equations, a series of experiments are performed, using real and synthetic textures. Each texture contains different combinations that represent the same motif with different perspectives and scales. These served to validate the equations proposed in this study. Moreover, in applications always present the same pattern of behavior for the equivalent equations of the original ones described by using GLCM. All the results were favorable to the use of SDH that presented a reduction of the complexity wich speeds up to 98.6% the computational time.

There is a great number of image applications, where the first step in the analysis is to associate the same point in two or more images that were taken from different viewpoints or at different times. This procedure is called image registration (IR). This work proposes a new efficient method for IR based on tensor calculus that overtakes the usual assumption of brightness consistency among the involved images. The main idea is to define transformations in serial images by using tensor behavior on coordinate transformations. It combines the simplicity and efficiency of the computer implementation with the correctness and elegance of the physic foundations. Moreover, it introduces a complete different way for management of this very important application of image processing: the use of the principal axis idea. The significant advantage of the proposed method is its promptitude and robustness for use in medical images in a way the as far as we known have never considered before. Experimental results in real examinations demonstrate that the proposed algorithm is highly accurate and outperforms a combination of the most updated IR approaches.

This paper presents the use of the Number of Texture Unit as a feature extractor for classification of breast images. The Number of Texture Unit served as the basis for the idealization of the Local Binary Pattern a technique that is widely used in facial recognition. We compared the proposed strategy with the Gray Level Co-occurrence Matrix which is the most used texture analysis technique in the literature. With this work we have been able to show that the combination of the two techniques of feature extraction improves the final result of classification. To perform the tests we used the Support Vectors Machine classifier and obtained a result of 96.15% Area Under the Curve (Receiver Operating Characteristic Curve).

Computation of texture information for image classification and pattern recognition is a highly complex activity. Much of this complexity is related to feature extractions where computational cost is related with the tonal levels and resolution of the images. One of the more used methods is the Gray Level Occurrence Matrix (GLCM) that present quadratic time complexity in relation to of gray levels. After the computation of such a matrix the Haralick descriptors are calculated from the GLCM and used for the next steps of classification for pattern recognition. However, the same descriptors can be calculated from another approach, with lower complexity and computational cost, presenting the same results as this work shows. This approach is based on the construction of the Sum and Difference Histograms (SDH). The computational complexity of this method is linear in relation to the amount of gray levels. This work, by examples, demonstrates the above mentioned statement.

Aging is a worldwide phenomenon and represents a growing concern for public health systems. In this context, neurodegenerative diseases like Alzheimer's Disease (AD) have a high prevalence among the elderly. Early diagnosis of AD allows early treatment and improves patient's quality of life. In this paper, we propose a clinical decision support system (CDSS) to aid physicians in diagnosis of AD and related disorders: Dementia (D) and Mild Cognitive Impairment (MCI). For each case, the system exhibits the most probable diagnosis, the most relevant health records and unobserved health records that could confirm such diagnosis. Moreover, the system has the ability to refine its decision model using the final diagnosis reported by physicians. The system can offer to physicians a friendly user interface designed for smartphones. The proposed decision support system is flexible and adaptable to different contexts, since it allows new neuropsychological tests to be included and its decision model to be adapted automatically. Clinical cases from Center for Alzheimer's Disease (CAD) of Federal University of Rio de Janeiro (UFRJ) were used as the training dataset for the proposed supervised learning method. Preliminary tests using clinical cases from Antônio Pedro University Hospital (HUAP) of Fluminense Federal University (UFF) showed that the proposed CDSS decision model achieves good performance (accuracy of 0.94 for D and 0.85 for MCI) by a computational method that evaluates several classifiers and selects the best for each mental disorder.

Early detection of thyroid anomalies decreases the chances of disease progression. Imaging examinations consist in an important tool in the diagnostic process. However, most of them are relatively expensive or can expose the patient to excessive radiation. Thermography is an interesting alternative in thyroid diseases diagnosis, especially in the detection of nodules, since some of them tend to present higher temperatures than normal tissues. Image processing techniques can be used to find regions that may indicate thyroid nodules. To select which one of these regions are in fact related to a nodule, a Convolutional Neural Network - CNN can be used. CNNs are widely used in clinical images classification, and some models have shown good results in this kind of problem. In this work, we present a methodology to identify thyroid nodules in thermograms by using simple image processing techniques and CNNs. Three CNNs were tested, the first one based in the GoogLeNet architecture, a second based in the AlexNet and a third one based in the VGG architecture. The GoogLeNet CNN yielded the highest accuracy (86.22%) followed by AlexNet (77.67%) and the VGG (74.96%).

This work proposes a variable neighbourhood search (FTS) that uses a fractal‐based local search primarily designed for images. Searching for specific content in images is posed as an optimisation problem, where evidence elements are expected to be present. Evidence elements improve the odds of finding the desired content and are closely associated to it in terms of spatial location. The proposed local search algorithm follows the fashion of a chain of triangles that engulf each other and grow indefinitely in a fractal fashion, while their orientation varies in each iteration. The authors carried out an extensive set of experiments, which confirmed that FTS outperforms state‐of‐the‐art metaheuristics. On average, FTS was able to locate content faster, visiting less incorrect image locations. In the first group of experiments, FTS was faster in seven out of nine cases, being >8% faster on average, when compared to the second best search method. In the second group, FTS was faster in six out of seven cases, and it was >22% faster on average when compared to the approach ranked second best. FTS tends to outperform other metaheuristics substantially as the size of the image increases.

In this study, we propose a system to classify thermal images of the breast considering the presence or absence of disease not only tumor and cancer, as most of previous works did. The proposal, at present stage, separates a breast among presenting benign tumor, cancer or being normal and shows how new diagnostic can be included in the data base when a number of proved case of a disease is known. The methodology begins by pre-processing images to be free of those regions that are not of interest for the work before execution of an algorithm for textural features computation, then a set of characteristics is calculated and they are ranged by machine learning classifier, that identity the clinical condition of the patient's breasts. For this classification, a Support Vector Machine (SVM) was used and trained based on already classified diagnosis from two data bases. Tests performed present accuracy above 90%, showing that the proposal is very promising. The project is in the phase of final adjustments and critical use by the medical community for the diagnostic aid. In future work integrations with other examinations is the aim.

The worldwide aging phenomenon is a growing concern. Alzheimer's disease (AD) has a high prevalence in the elderly. In this paper, we present a clinical decision support system for aiding the diagnosis of AD and related disorders. We describe system's main components and architecture, which is based on a mobile web-based platform. Its predictive model is based on Bayesian networks designed considering AD diagnosis criteria, trained and tested with the patient database of the Center for Alzheimer's Disease and Related Disorder at the Institute of Psychiatry of the Federal University of Rio de Janeiro, Brazil. Patient database attributes are composed by predisposal factors, demographic data, assessment scales, symptoms and signs. When the system indicates a patient diagnosis, it provides: the most probable diagnosis, health data that lead to such diagnosis and, in case of low certainty factor, unobserved health data that should be collected to confirm or refuse the initial diagnostic hypothesis. Preliminary usability tests indicate potential use of the system in clinical practice.

The Computational Theory of Perceptions (CTP) provides capabilities for linguistic summarization of data and it aims the description of patterns emerging from these data by means of linguistic expressions. This technique is particularly well suited in applications where there is the need of understanding the information at different levels of expertise and/or when intense human-computer interaction is required. In this paper, we present a CTP-based system able to generate valuable linguistic reports from findings in breast image mammograms using the BI-RADS radiology standard. The implemented framework uses data obtained through the fusion of information provided by different medical experts on the same mammography. Then, our system automatically produces a collection of valid sentences describing: the breast lesion findings (i.e. features of masses or calcifications), the BI-RADS category rating for the analyzed mammography and the recommendations for patients. The system validation by several medical specialists has produced promising results, which make it useful for its practical deployment.

This work incorporates the AES cryptography algorithm, to improve the hidden data security in two methodologies for steganography: the genetic algorithm and path relinking. It also combines them proposing a new hybrid approach that outperforms the LSB (least significant bits) substitution technique presented in works cited in the literature concerning the quality of a stego image. It improves the possibility of hiding data inside color images significantly, increasing the space available for information by more than three times when compared to the usual steganography approach used by grayscale images. Moreover all types of digital information from text and compressed files to even executable programs can be hidden inside the cover image. This considerably increases the scope of application of the technique for transmitting information inside a typical image, hiding the data from intruders.

This work proposes a context-aware middleware for medical workflow organization and efficiency improvement. In hospitals, laboratories and teleradiology companies, each physician or technician is specialized in a specific kind of diagnosis or analysis. Therefore, certain types of medical images are often forwarded to a certain physician or a certain group. This forwarding is time consuming. That is, repeatedly deciding who would be the best physician, whether he is available at a certain moment given a certain context is exhaustive and may be very inefficient. Thus, the proposed middleware has the ability to process and collect data from images analyzed by each medical staff. Based on the collected data and current clinical context, the middleware is able to infer who would be the best fit staff to receive a certain incoming medical image.

This paper addresses the image registration problem applying genetic algorithms. The image registration's objective is the definition of a mapping that best match two set of points or images. In this work the point matching problem was addressed employing a method based on nearest-neighbor. The mapping was handled by affine transformations. Experiments were conducted using three 2D synthetic point-sets with different affine transformations and noise. The results were compared against other optimization techniques. The similarity of two point-sets is measured using the Euclidean distance between matched points.

This work presents an application of software engineering to fabric inspection. An inspection system has been developed for textile industries that aims automatic failure detection. Such as wood, paper and steel industries, this environment has particular characteristics in which surface defect detection is used for quality control. This system combines concept from software engineering and decision support. Detection of defects within the inspected texture is performed in a first step acquiring images by CCD cameras, then extracting texture features and, finally by classifiers being trained a priori on database of defective and non-defective samples. The extracted data depend on the type of method selected for image analysis. The used types are based on segmentation or fractal dimension. Two usual segmentation techniques were adapted and improved. A new algorithm was developed to calculate efficiently fractal dimension of textures. Experiments show the accuracy and applicability of the proposed techniques for a real factory environment.

One important applications of fractals is the field of image texture analysis. The main aspect of fractal geometry used in such application is the concept of fractal dimensions to characterize the texture scaling behavior. However, this identification with fractal makes sense only within certain limits. Moreover sets with the same fractal dimension may differ substantially in their structure. One proposition to handle this is to describe the set not only by one fractal dimension, but by a set of dimensions with their properties. We propose the characterization of multifractal set by the local Hausdorff dimension and two local box-counting dimensions. A new idea for Hausdorff dimension calculation of images is also presented.

Many texture classification schemes require an excessively large image area for texture analysis, use a large number of features to represent each texture or are computationally very demanding. In this paper we describe a segmentation method using color and fractal dimension for real time texture classification. The box-counting approach is used to estimate the fractal dimension (FD). A seed block which embodies information about color features and FD is used by a region growing method. Experimental results indicate that the proposed method is promising for color texture segmentation. This scheme is computationally very efficient and it is suited for texture image recognition.