Figure 4 - uploaded by Juan Andrade Cetto
Content may be subject to copyright.
Positive and negative samples.
Source publication
We present a framework for object detection that is invariant to object translation, scale, rotation, and to some degree, occlusion, achieving high detection rates, at 14 fps in color images and at 30 fps in gray scale images. Our approach is based on boosting over a set of simple local features. In contrast to previous approaches, and to efficient...
Citations
... The learning rate for shrinkage in boosting (LR) is set as 0.1 which is a popular choice and have been adopted in various publications (Hu et al., 2016, Paisitkriangkrai et al., 2014, Mounce et al., 2017. A Haar-like feature implementation by (Villamizar et al., 2006) is adopted with adaptations. ...
Breast mass detection and segmentation are challenging tasks due to the fact that breast masses vary in size and appearance. In this work, we present a simultaneous detection and segmentation scheme for mammographic lesions that is constructed in a sifting architecture. It utilizes a novel region candidate selection approach and cascaded learning techniques to achieve state-of-the-art results while handling a high class imbalance. The region candidates are generated by a novel multi-scale morphological sifting (MMS) approach, where oriented linear structuring elements are used to sieve out the mass-like objects in mammograms including stellate patterns. This method can accurately segment masses of various shapes and sizes from the background tissue. To tackle the class imbalance problem, two different ensemble learning methods are utilized: a novel self-grown cascaded random forests (CasRFs) and the random under-sampling boost (RUSBoost). The CasRFs is designed to handle class imbalance adaptively using a probability-ranking based under-sampling approach, while RUSBoost uses a random under-sampling technique. This work is evaluated on two publicly available datasets: INbreast and DDSM BCRP. On INbreast, the proposed method achieves an average sensitivity of 0.90 with 0.9 false positives per image (FPI) using CasRFs and with 1.2 FPI using RUSBoost. On DDSM BCRP, the method yields a sensitivity of 0.81 with 3.1 FPI using CasRFs and with 2.9 FPI using RUSboost. The performance of the proposed method compares favorably to the state-of-the-art methods on both datasets, especially on highly spiculated lesions.
... On the other hand, there exist methods based on local features. In this case, objects are represented via their edges, colour or corner cues [27], [28], [29]; steerable filters [30]; haarlike features [31]; or scale-invariant descriptors (e.g. SIFT, SURF) [32], [33]. ...
Technological advances are being made to assist humans in performing ordinary tasks in everyday settings. A key issue is the interaction with objects of varying size, shape, and degree of mobility. Autonomous assistive robots must be provided with the ability to process visual data in real time so that they can react adequately for quickly adapting to changes in the environment. Reliable object detection and recognition is usually a necessary early step to achieve this goal. In spite of significant research achievements, this issue still remains a challenge when real-life scenarios are considered. In this article, we present a vision system for assistive robots that is able to detect and recognize objects from a visual input in ordinary environments in real time. The system computes color, motion, and shape cues, combining them in a probabilistic manner to accurately achieve object detection and recognition, taking some inspiration from vision science. In addition, with the purpose of processing the input visual data in real time, a graphical processing unit (GPU) has been employed. The presented approach has been implemented and evaluated on a humanoid robot torso located at realistic scenarios. For further experimental validation, a public image repository for object recognition has been used, allowing a quantitative comparison with respect to other state-of-the-art techniques when realworld scenes are considered. Finally, a temporal analysis of the performance is provided with respect to image resolution and the number of target objects in the scene.
... 2) Effect of using multi-sequence MRI and Deep Networks Architecture: Fig. 7b presents the mean landmark-to-surface distance of the AVP segmentation framework, PAScAL, for employing intensity values from a single MRI sequence (T1), multiple MRI sequences, and multiple MRI sequences with hand-crafted Haar-like features [33], and the deep learned (SAE) features. An average landmark-to-surface distance of 2.25 ± 0.98 mm (p-value< 0.001) was obtained with T1weighted sequences only, 1.86 ± 0.86 mm (p-value< 0.001) with T1-weighted, T2-weighted, and FLAIR MRI sequences, 1.28 ± 0.53 mm (p-value< 0.001) with hand-crafted Haar-like features, and 0.66 ± 0.31 mm for MRI sequences with SAE features. ...
Analysis of cranial nerve systems, such as the anterior visual pathway (AVP), from MRI sequences is challenging due to their thin long architecture, structural variations along the path, and low contrast with adjacent anatomic structures. Segmentation of a pathologic AVP (e.g. with low-grade gliomas) poses additional challenges. In this work, we propose a fully automated partitioned shape model segmentation mechanism for AVP steered by multiple MRI sequences and deep learning features. Employing deep learning feature representation, this framework presents a joint partitioned statistical shape model able to deal with healthy and pathological AVP. The deep learning assistance is particularly useful in the poor contrast regions, such as optic tracts and pathological areas. Our main contributions are: (1) a fast and robust shape localization method using conditional space deep learning, (2) a volumetric multiscale curvelet transform- based intensity normalization method for robust statistical model, and (3) optimally partitioned statistical shape and appearance models based on regional shape variations for greater local flexibility. Our method was evaluated on MRI sequences obtained from 165 pediatric subjects. A mean Dice similarity coefficient of 0.779 was obtained for the segmentation of the entire AVP (optic nerve only=0.791) using the leave-one-out validation. Results demonstrated that the proposed localized shape and sparse appearance-based learning approach significantly outperforms current state-of-the-art segmentation approaches and is as robust as the manual segmentation.
... Yet, many recent methods have shown a remarkable success when are used in conjunction with machine learning techniques such as boosting[10] [14] [22] [26] [28] or Support Vector Machines (SVMs) [2] [4] [9] [15] [18]. However, these methods have been effectively used mostly for standard datasets [1] [3] [5] [11] for which the objects only appear in a relatively reduced number of poses [7] [20] [24]. ...
... The problem of detecting object categories in images is known to be very challenging and needs to address several issues such as large intra-class object variations, changes in object pose, cluttered backgrounds or illumination changes. Yet, many recent methods have shown a remarkable success when are used in conjunction with machine learning techniques such as boosting[10, 14, 22, 26, 28] or Support Vector Machines (SVMs) [2, 4, 9, 15, 18]. However, these methods have been effectively used mostly for standard datasets [1, 3, 5, 11] for which the objects only appear in a relatively reduced number of poses [7, 20, 24]. ...
We present a new approach for building an efficient and robust classifier for the two class problem, that localizes objects that may appear in the image under different orientations. In contrast to other works that address this problem using multiple classifiers, each one specialized for a specific orientation, we propose a simple two-step approach with an estimation stage and a classification stage. The estimator yields an initial set of potential object poses that are then validated by the classifier. This methodology allows reducing the time complexity of the algorithm while classification results remain high. The classifier we use in both stages is based on a boosted combination of Random Ferns over local histograms of oriented gradients (HOGs), which we compute during a preprocessing step. Both the use of supervised learning and working on the gradient space makes our approach robust while being efficient at run-time. We show these properties by thorough testing on standard databases and on a new database made of motorbikes under planar rotations, and with challenging conditions such as cluttered backgrounds, changing illumination conditions and partial occlusions.
... ! !Figure 2. Approximating a half-disc with a rectangle We then compute an integral image for each bin of the histogram, similarly to [17] . Complicating the use of integral images in this context is the fact that integral images can only compute sums of rectangles, whereas we need to compute sums of rotated rectangles. ...
Image contour detection is fundamental to many image analysis applications, including image segmentation, object recognition and classification. However, highly accurate image contour detection algorithms are also very computationally intensive, which limits their applicability, even for offline batch processing. In this work, we examine efficient parallel algorithms for performing image contour detection, with particular attention paid to local image analysis as well as the generalized eigensolver used in Normalized Cuts. Combining these algorithms into a contour detector, along with careful implementation on highly parallel, commodity processors from Nvidia, our contour detector provides uncompromised contour accuracy, with an F-metric of 0.70 on the Berkeley Segmentation Dataset. Runtime is reduced from 4 minutes to 1.8 seconds. The efficiency gains we realize enable high-quality image contour detection on much larger images than previously practical, and the algorithms we propose are applicable to several image segmentation approaches. Efficient, scalable, yet highly accurate image contour detection will facilitate increased performance in many computer vision applications.
... Unlike Laptev's work, our boosted features are not computed in random locations but computed exhaustively over the whole image with the aim to determine which image locations are human parts and which ones are background. The training is carried out using the well-known Adaboost algorithm that has successful results in object detection [6,7,8]. Given that pedestrian HoGs are corrupted by background, we propose HoG-based features instead of whole HoG descriptors in order to concentrate on HoG parts with high reliability. ...
The present paper addresses pedestrian detection using local boosted features that are learned from a small set of training
images. Our contribution is to use two boosting steps. The first one learns discriminant local features corresponding to pedestrian
parts and the second one selects and combines these boosted features into a robust class classifier. In contrast of other
works, our features are based on local differences over Histograms of Oriented Gradients (HoGs). Experiments carried out to
a public dataset of pedestrian images show good performance with high classification rates.
... Boosting algorithms are very well known methods for fast object detection which are based on building robust classifiers from simple (weak) features [4, 5]. We follow the framework addressed in [6], but based on contours instead of intensity images. The use of contour images allows the use of inner and outer object contours to perform robust detection without the drawback of background. ...
In this work a new robust color and contour based object detection method in images with varying shadows is presented. The method relies on a physics-based contour detector that emphasizes material changes and a contour-based boosted classifier. The method has been tested in a sequence of outdoor color images presenting varying shadows using two classifiers, one that learns contour object features from a simple gradient detector, and another that learns from the photometric invariant contour detector. It is shown that the detection performance of the classifier trained with the photometric invariant detector is significantly higher than that of the classifier trained with gradient detector. Peer Reviewed
... Other extensions have been proposed to calculate other local properties efficiently. Villamizar et al [5] and Porikli [6] developed the Integral Histogram, with which is possible to compute rapidly any local histogram independently of its size and location. ...
... The work reported here introduces a novel multiscale unidimensional histogram representation based on a linear combination of Haar features, that follows the spirit of other typical feature sets learned via Boosting. These histograms are efficiently computed using our previously reported integral histogram image [5] and we compare on its use for object detection against the Swain and Ballard histogram intersection metric. ...
... Once computed this image representation, any one of the local features can be computed at any location and scale in constant time [4]. Extending the idea of having cumulative data at each pixel in the integral image, we have proposed to store on it the histogram data instead of intensity sums [5]. The integral histogram stores intensity level histograms which, once constructed, allow for the computation of histogram within a rectangular area in constant time. ...
In this article, scale and orientation invariant object detection is performed by matching intensity level histograms. Unlike other global measurement methods, the present one uses a local feature description that allows small changes in the histogram signature, giving robustness to partial occlusions. Local features over the object histogram are extracted during a Boosting learning phase, selecting the most discriminant features within a training histogram image set. The Integral Histogram has been used to compute local histograms in constant time. Peer Reviewed
... The type of weak classifier features used in [5] are very simple template matching masks, that would presumibly fail if sample objects are to be found at different orientations than as trained. In this work we investigate on the use of similar multiclass feature selection, but with keen interest in fast computation of orientation invariant weak classifiers [6] for multiclass rotation invariant object recognition. In [2], Viola introduced the integral image for very fast feature evaluation. ...
... In [6] we realized that filter response to Haar masks can be not only be computed efficiently with an integral image scheme; but also, that such masks can be approximately rotated with some simplifications of the Gaussian steerable filter. Thus, allowing for fast computation of rotation invariant filter responses as week classifiers. ...
... and, the steered 2nd order Gaussian filter can be obtained withFigure 2. Convolving with Gaussian kernels is a time consuming process. Instead, we propose in [6] to approximate such filter response by convolving with the Haar basis with the objective of using the integral image. Thus, we approximate the oriented first derivative response with ...
We present a framework for object recognition based on sim- ple scale and orientation invariant local features that when combined with a hierarchical multiclass boosting mechanism produce robust clas- sifiers for a limited number of object classes in cluttered backgrounds. The system extracts the most relevant features from a set of training sam- ples and builds a hierarchical structure of them. By focusing on those features common to all trained objects, and also searching for those fea- tures particular to a reduced number of classes, and eventually, to each object class. To allow for efficient rotation invariance, we propose the use of non-Gaussian steerable filters, together with an Orientation Integral Image for a speedy computation of local orientation.
... Pero los descriptores SIFT no son la única investigación en el campo de la extracción de características invariantes. En la UPC-CSIC de Barcelona [43], también se ha realizado un estudio sobre características invariantes a la rotación desde otro enfoque. La principal diferencia es la selección de características multiclase pero con un fuerte interés en el cálculo rápido de clasificadores débiles invariantes a orientación [43] con el fin de conseguir un sistema de reconocimiento de múltiples objetos invariantes a rotación. ...
... En la UPC-CSIC de Barcelona [43], también se ha realizado un estudio sobre características invariantes a la rotación desde otro enfoque. La principal diferencia es la selección de características multiclase pero con un fuerte interés en el cálculo rápido de clasificadores débiles invariantes a orientación [43] con el fin de conseguir un sistema de reconocimiento de múltiples objetos invariantes a rotación. Para la selección de características, las que mejor discriminan un objeto se obtienen gracias a la convolución de muestras de imágenes positivas con un conjunto simplificado de funciones bases wavelets a diferentes escalas y orientaciones. ...
La proliferación de diferentes fuentes de información de texto, imágenes y video en formato digital hace interesante la investigación de tecnologías de procesado de datos que puedan aplicarse independientemente del tipo de información permitiendo, eventualmente, un procesado integrado de múltiples fuentes de una manera que cupiera denominar semántica; por su nivel de abstracción y su potencial acercamiento a la forma en que el ser humano maneja la información. Este proyecto fin de carrera hace un análisis preliminar del empleo del análisis espectral de grafos definidos a partir de imágenes con el propósito anterior. Los resultados de esta investigación muestran que los espectros de los grafos definidos son mayormente degenerados, imposibilitando su uso para la categorización de imágenes, y que es necesaria una reorientación en la definición de los mismos.
