Caption Region Research Articles

Learning Object Context for Dense Captioning

Dense captioning is a challenging task which not only detects visual elements in images but also generates natural language sentences to describe them. Previous approaches do not leverage object information in images for this task. However, objects provide valuable cues to help predict the locations of caption regions as caption regions often highly overlap with objects (i.e. caption regions are usually parts of objects or combinations of them). Meanwhile, objects also provide important information for describing a target caption region as the corresponding description not only depicts its properties, but also involves its interactions with objects in the image. In this work, we propose a novel scheme with an object context encoding Long Short-Term Memory (LSTM) network to automatically learn complementary object context for each caption region, transferring knowledge from objects to caption regions. All contextual objects are arranged as a sequence and progressively fed into the context encoding module to obtain context features. Then both the learned object context features and region features are used to predict the bounding box offsets and generate the descriptions. The context learning procedure is in conjunction with the optimization of both location prediction and caption generation, thus enabling the object context encoding LSTM to capture and aggregate useful object context. Experiments on benchmark datasets demonstrate the superiority of our proposed approach over the state-of-the-art methods.

Video Caption Extraction Using Spatio-Temporal Slices

Captions in videos play an important role for video indexing and retrieval. In this paper, we propose a novel algorithm to extract multilingual captions from video. Our approach is based on the analysis of spatio-temporal slices of video. If the horizontal (or vertical) scan line contains some pixels of caption region then the corresponding spatio-temporal slice will have bar-code like patterns. By integrating the structure information of bar-code like patterns in horizontal and vertical slices, the spatial and temporal positions of video captions can be located accurately. Experimental results show that the proposed algorithm is effective and outperforms some existing techniques.

Caption Region Detection in Video Images by Improved Corner Detector

Caption region detection is significant for text segmentation and recognition in video images. In this paper, caption region detection includes the following three steps: text corner detection, candidate text region generation, non-text region elimination. Firstly, an improved Harris corner detection algorithm based on second difference, called sHarris, is proposed to detect text corners. sHarris detector is effective to detect text corners existing in video image. Secondly, morphologic operations are utilized to generate candidate text regions after text corner points are detected. Finally, non-text regions are eliminated from candidate text regions by some validation rules. In the end of this paper, experiments are used to analyze and evaluate the detection results of caption regions, which indicate the effectiveness of the proposed method.

Eye movements while viewing narrated, captioned, and silent videos

Videos are often accompanied by narration delivered either by an audio stream or by captions, yet little is known about saccadic patterns while viewing narrated video displays. Eye movements were recorded while viewing video clips with (a) audio narration, (b) captions, (c) no narration, or (d) concurrent captions and audio. A surprisingly large proportion of time (>40%) was spent reading captions even in the presence of a redundant audio stream. Redundant audio did not affect the saccadic reading patterns but did lead to skipping of some portions of the captions and to delays of saccades made into the caption region. In the absence of captions, fixations were drawn to regions with a high density of information, such as the central region of the display, and to regions with high levels of temporal change (actions and events), regardless of the presence of narration. The strong attraction to captions, with or without redundant audio, raises the question of what determines how time is apportioned between captions and video regions so as to minimize information loss. The strategies of apportioning time may be based on several factors, including the inherent attraction of the line of sight to any available text, the moment by moment impressions of the relative importance of the information in the caption and the video, and the drive to integrate visual text accompanied by audio into a single narrative stream.

시·공간 정보를 이용한 동영상의 인공 캡션 검출

동영상에 포함되는 인공 캡션은 영상과 관계있는 의미정보를 포함한다. 이러한 영상을 표현하는 정보를 이용하기 위해 캡션을 추출하는 연구는 근래에 들어 활발히 진행되고 있다. 기존 방법들은 대부분 정지영상에서 캡션을 검출하였다. 하지만 동영상의 경우에는 유용한 시간정보가 있다. 따라서 본 연구는 이러한 시간정보를 사용한 캡션영역 검출방법을 제안한다. 먼저, 캡션후보영역 검출을 위해 문자출현맵을 생성하고, 후보영역 매칭 과정에서 지속후보영역을 검출한다. 검출된 지속후보영역의 소멸성 검사를 통해 캡션의 소멸 여부를 검출하고 소멸된 캡션 일 경우 시 공간정보에 의한 병합과정을 통해 캡션후보영역을 결정한다. 마지막으로 결정된 캡션후보영역을 검증하기 위하여 에지 방향 히스토그램을 이용한 신경망 인식기를 통하여 최종캡션영역을 검출한다. 실험을 위해 다양한 크기와 형태, 위치의 캡션을 포함하는 동영상에 대해 영역검출의 성능을 평가하고자 Recall과 Precision을 이용하여 제안하는 방법의 영역검출에 대한 효율성을 입증한다. The artificial captions appearing in videos include information that relates to the videos. In order to obtain the information carried by captions, many methods for caption extraction from videos have been studied. Most traditional methods of detecting caption region have used one frame. However video include not only spatial information but also temporal information. So we propose a method of detection caption region using temporal and spatial information. First, we make improved Text-Appearance-Map and detect continuous candidate regions through matching between candidate-regions. Second, we detect disappearing captions using disappearance test in candidate regions. In case of captions disappear, the caption regions are decided by a merging process which use temporal and spatial information. Final, we decide final caption regions through ANNs using edge direction histograms for verification. Our proposed method was experienced on many kinds of captions with a variety of sizes, shapes, positions and the experiment result was evaluated through Recall and Precision.

Story-related caption detection and localization in news video

We propose a method to detect and localize story-related subject captions in news video. Most caption detection and localization algorithms attempt to detect as many captions as possible; however, a news picture may contain many types of captions that are unrelated to the story. To facilitate fast and accurate access to news video content, a method for detecting and localizing the story-related caption is necessary. This paper addresses two problems in texture-based caption detection and localization: the time-consuming computation of features, and the clutter of caption detection results. We address these problems by first identifying the subject caption region based on the frequency of text occurrence. Then, we detect the subject caption frame as it first appears onscreen. Finally, the texture-based caption localization procedure is performed on the subject caption region in subject captions' beginning frames. Using this method decreases the computation time significantly. Additionally, the unrelated types of text are also filtered out, and only the subject caption is detected and localized. Experimental results show that the proposed method can quickly and robustly detect subject captions from news video.

Caption analysis and recognition for building video indexing systems

In this paper, we propose several methods for analyzing and recognizing Chinese video captions, which constitute a very useful information source for video content. Image binarization, performed by combining a global threshold method and a window-based method, is used to obtain clearer images of characters, and a caption-tracking scheme is used to locate caption regions and detect caption changes. The separation of characters from possibly complex backgrounds is achieved by using size and color constraints and by cross examination of multiframe images. To segment individual characters, we use a dynamic split-and-merge strategy. Finally, we propose a character recognition process using a prototype classification method, supplemented by a disambiguation process using support vector machines, to improve recognition outcomes. This is followed by a postprocess that integrates multiple recognition results. The overall accuracy rate for the entire process applied to test video films is 94.11%.