Learning and matching multiscale template descriptors for real-time detection, localization and tracking

Lee, Taehee; Soatto, Stefano

doi:10.1109/cvpr.2011.5995453

Cited by 21 publications

(26 citation statements)

References 20 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…This is the fundamental link that ties together sensing and control: Without control, no (time-and risk-optimal) sensing can be performed in the presence of line-of-sight and quantization phenomena, and clearly without sensing no (feedback) control could be performed. Follow-up work includes the course notes [68], a characterization of detachable object detection, [7,6], a characterization of the controlled recognition bounds [70], of sufficient exploration [62], and the optimal descriptor TST/BTD [49]. .…”

Section: Summary and Discussionmentioning

confidence: 99%

Actionable Information in Vision

Soatto

2013

Machine Learning for Computer Vision

Self Cite

View full text Add to dashboard Cite

Summary. A notion of visual information is introduced as the complexity not of the raw images, but of the images after the effects of nuisance factors such as viewpoint and illumination are discounted. It is rooted in ideas of J. J. Gibson, and stands in contrast to traditional information as entropy or coding length of the data regardless of its use, and regardless of the nuisance factors affecting it. The non-invertibility of nuisances such as occlusion and quantization induces an "information gap" that can only be bridged by controlling the data acquisition process. Measuring visual information entails early vision operations, tailored to the structure of the nuisances so as to be "lossless" with respect to visual decision and control tasks (as opposed to data transmission and storage tasks implicit in communications theory). The definition of visual information suggests desirable properties that a visual representation should possess to best accomplish vision-based decision and control tasks.

show abstract

Section: Summary and Discussionmentioning

confidence: 99%

Actionable Information in Vision

Soatto

2013

Machine Learning for Computer Vision

Self Cite

View full text Add to dashboard Cite

show abstract

“…However in many cases, the use of a more complicated model for similarity measurement results in higher computational cost. Further, sliding window search is also a costly stage for such methods [14]. While there exist known algorithms for fast search of object instances in an image using branch-and-bound techniques [13], in our particular problem, methods of this type have two crucial limitations.…”

Section: Introductionmentioning

confidence: 99%

“…Then the normalized density score with respect to the spatially, uniformly distributed response is shown in Eqn. (14).…”

Section: Matching Detectionmentioning

confidence: 99%

“…According to the way similarity measurements are performed, such template matching methods can roughly be classified into two groups: 1) patch matching schemes, such as the sum of absolute difference (SAD) [1], the sum of squared difference (SSD) [5], or cross correlation (XCORR) [27], where the similarity measurement relies directly on pixel information from the patch of interest; and 2) feature matching schemes, such as invariant features [3,15] and bags of features [14,16,22,24], where similarity measurement relies on features describing the template and the frame. Patch matching methods are not robust, especially when noise, skew, or errors occur [25].…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Wavelet Band-pass Filters for Matching Multiple Templates in Real-time

Wu¹,

Noonan²,

Prasad³

et al. 2011

Procedings of the British Machine Vision Conference 2011

View full text Add to dashboard Cite

Most existing methods for template matching require computationally demanding searches and may not be appropriate for matching multiple templates. We present a novel real-time solution by matching a number of subtemplates with an input image, instead of using the entire template. Each template class is first learned and depicted offline as a set of wavelet-based directional band-pass filters, each of which allows only the most salient wavelet frequencies (subtemplates) to pass. As a result, matching a template becomes equivalent to the problem of filtering the wavelet transformed image using a set of band-pass filters. Finally, the region with the highest response density after filtering is considered as the detection. The computational complexity of this matching method is 1/50 of optimal correlation based SSD [5]and 1/10 of orthogonal Haar transform based SSD [20]. Further, the use of subtemplates enables detection even in the presence of skew and rotation. We present experimental results that demonstrate the capacity of our system for detection with a total of 10 different logo classes in broadcast videos.

show abstract

“…Table 1: 30 kinds of food ingredients in the data set. types ingredients fish (5) tuna, squid, octopus, shrimp, salmon meat (6) beef, pork, chicken, minced meat, sausage, ham vegetable mushroom, potato, eggplant, carrot, radish, (13) tomato, cucumber, cabbage, green onions, onion, Chinese cabbage, lettuce, Shiitake mushroom fruit (6) apple, strawberry, pineapple, orange, banana, grapefruit…”

Section: Data Collection and Experimental Settingmentioning

confidence: 99%

Real-time mobile recipe recommendation system using food ingredient recognition

Maruyama

Kawano

Yanai

2012

Proceedings of the 2nd ACM International Workshop on Interactive Multimedia on Mobile and Portable Devices

View full text Add to dashboard Cite

In this paper, we propose a mobile cooking recipe recommendation system employing object recognition for food ingredients such as vegetables and meats. The proposed system carries out object recognition on food ingredients in a real-time way on an Android-based smartphone, and recommends cooking recipes related to the recognized food ingredients. By only pointing a built-in camera on a mobile device to food ingredients, the user can obtain a recipe list instantly. As an object recognition method, we adopt bagof-features with SURF and color histogram extracted from multiple images as image features and linear SVM with the one-vs-rest strategy as a classifier. We built 30 kinds of food ingredient short video database for experiments. With this database, we achieved the 83.93% recognition rate within the top six candidates. In the experiment, we made user study by comparing mobile recipe recommendation systems with/without ingredient recognition.

show abstract

Learning and matching multiscale template descriptors for real-time detection, localization and tracking

Cited by 21 publications

References 20 publications

Actionable Information in Vision

Actionable Information in Vision

Wavelet Band-pass Filters for Matching Multiple Templates in Real-time

Real-time mobile recipe recommendation system using food ingredient recognition

Contact Info

Product

Resources

About