<title>Ground truth for training and evaluation of automatic main subject detection</title>

Abstract

A consumer photograph, or snapshot, is a medium for conveying to a viewer, one's interest in one or more main subjects. A methodology is presented for collecting ground truth data useful for training and evaluating algorithms designed to automatically detect the main subject of a consumer photograph. For a database of 100 images, 16 observers provided polygonal approximations to the image areas that comprise the main subject. Results from all observers are combined to form a truth image that is considered the ideal result of a main subject detector and is analyzed to determine features for main subject detection (MSD) . The collected ground truth shows substantial agreement among third-party observers. It also supports conventional wisdom regarding the likely locations of main subjects and the value of "people" detection as a cue for main subject detection. Training data is created from the truth images for an MSD framework involving image segmentation, feature detection, and probabilistic reasoning. A proposed method for generating region-based training data can be used to retrain a reasoning engine as segmentation algorithms improve, without further observer involvement. Although the subject matter for consumer photographs ranges from sweeping landscapes to close portraits, identification of the main subject is a meaningful task.

An unsupervised approach to determination of main subject regions in images with low depth of field

“…The sharp regions consisting of sharp blocks are the main subject regions. for j = 1 to n do 5: Compute the DCT coefficient T of B ij 6: Compute β (2,2)i,j of T 7: if β (2,2)i,j > ξ max then 8: β (2,2)i,j = ξ max 9: end if 10: if β (2,2)i,j < θ DCT then if Dist(β (2,2)i,j , µ C1 ) > Dist(β (2,2)i,j , µ C2 ) then 14: if Dist(β (2,2)i,j , µ C1 ) < Dist(β (2,2)i,j , µ C2 ) then …”

“…But this method need construct a training set, and require the expert to distinguish the two different regions. Luo, Etz, Singhal [2] [3] proposed a Bayes neural network to detect the main subject region, and it is also a supervised learning method. Li,Wang et al [4] [5] analyze the statistics of the high-frequency wavelet coefficients to segment the focused regions in an image, then detect the main subject regions.…”

A computational approach to determination of main subject regions in photographic images

“…The central zone of the image is then classified as the main subject and all the other zones are classified as background. Since our application consists of consumer-type photographic images, there is strong evidence that main subject regions are located more towards the center of the image rather than towards the borders [10]. The performance of the Czone predictor (compared to the ground truth) sets the baseline for the Bayesian network and neural network classifiers.…”

“…We define centrality by computing the integral of a probability density function (PDF) of main subject location over the area of a given region. The PDF is derived from the ground truth data [10]. In doing so, every pixel of a given region, not just the centroid, contributes to the centrality measure of the region to a varying degree depending on its location.…”

An image-set for identifying multiple regions/levels of interest in digital images

“…Testing and performance evaluation of all these algorithms requires MROI and MLOI database. An experiment for developing an image-set and its ground-truth for training and evaluation of automatic main subject detection (people detection) algorithms were proposed in [2]. The authors used 100 images of consumer photographs to develop ground-truth of salient regions.…”